{ "data": { "a": { "slug": "decision-3", "name": "Decision 3.0", "vendor": "vLLM Semantic Router project", "vendorUrl": "https://vllm-sr.ai", "kind": "model", "category": "decision-models", "summary": "Decision 3.0 is a family of open-weight multimodal decision models from the vLLM Semantic Router team, 0.6B to 27B, under Apache-2.0. They answer choice, yes-or-no and score questions about text, images and video with a probability per option.", "url": "https://www.anchorterminal.com/tools/decision-3", "markdownUrl": "https://www.anchorterminal.com/tools/decision-3.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/decision-3.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/decision-3.json", "dns": "decision-3.grade.anchorterminal.com", "repo": "https://github.com/vllm-project/semantic-router", "license": "Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases", "transports": [ "http" ], "packages": [ { "registry": "pypi", "name": "vllm-sr" } ], "auth": "none", "authNotes": "No account. The weights are public and ungated on Hugging Face. The bundled `d3_server.py` binds to 127.0.0.1 and is open unless `DECISION_API_KEY` is set, after which it requires `Authorization: Bearer \u003ckey\u003e` and answers 401 otherwise. `GET /health` stays open. The router's model runtime has no authentication.", "pricing": "free", "pricingNotes": "Free and open source, with nothing to buy. Since mid-October the authors also run a free API at decision.vllm-sr.ai with per-key rate limits and daily quotas and no account, which logs requests and responses and may train models on them. Hardware is the owner's cost (checked 2026-11-02).", "priceSummary": "Free · OSS", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402. Decision 3.0 is software the owner runs, and neither the bundled server nor the router's model runtime has a payment route (checked 2026-10-12).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-12" }, "docsUrl": "https://huggingface.co/collections/vllm-sr/decision-30", "openapi": "https://raw.githubusercontent.com/vllm-project/semantic-router/main/src/model-runtime/vllm_srun/api/openapi.yaml", "capabilities": [ "inference.decision" ], "tags": [ "model", "open-source", "open-weights", "self-hosted", "local", "free", "python", "multimodal", "openapi" ], "lastRelease": "2026-10-30", "graded": true, "anchor": { "graded": true, "score": 66.7, "grade": "B", "agentReady": false, "rank": 460, "ranked": true, "rankOf": 1524, "categoryRank": 11, "methodology": "0.8", "run": "2026-10-01", "scores": { "ergonomics": 72, "maintenance": 74, "payments": 90, "reliability": 61, "schema": 70, "security": 53, "transparency": 51 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-02" }, "negative": 0, "verdict": "Open Apache-2.0 decision models, now eight sizes, that read images and video as well as text and answer typed questions in one call with a probability per option. vLLM 0.33.0 serves them natively. Three 27B models arrived in 14 days with no deprecation policy, loading with `transformers` needs `trust_remote_code`, and the authors' free API may train on requests.", "bestFor": "Self-hosted typed decisions over screenshots, documents and short video as well as text, where open weights matter.", "strengths": [ "Text, images and video in one request, with every question seeing all of them", "Apache-2.0 weights on Apache-2.0 bases, ungated on Hugging Face, with a SHA-256 manifest the runtime checks at load", "Hub releases tagged v3.0.0 to v3.1.3 on the first six sizes, so a revision can be pinned", "vLLM 0.33.0 serves d3 natively, and the authors publish NVFP4, INT4 and FP8 builds of the 27B models", "The bundled server takes only data URLs within stated size caps, so it fetches nothing from the network" ], "weaknesses": [ "Three 27B models (d3, d3-turbo, d3-ultra) in 14 days, with no deprecation policy or notice period", "No calibration figures on the cards, though the category compares models on calibration", "The authors' free API logs requests and responses and may train models on them, with no header to turn it off", "Loading with `transformers` needs `trust_remote_code=True`, and the router's Engine mode needs its development channel", "No training data, training code or changelog published with the models" ], "agentNotes": [ "Pin a tag or commit with `revision=` when loading from the Hub. d3 had 11 tagged releases, and new 27B models keep replacing the newest", "Send images and videos to `d3_server.py` as base64 data URLs only. The server refuses paths and http URLs with 422", "Set `DECISION_API_KEY` before binding the bundled server beyond 127.0.0.1. It is open otherwise", "Turn on `permutation_average=True` when routing must not depend on option order. It is off by default", "Expect one request at a time from the bundled server. Use the router's model runtime for queued serving", "Don't send sensitive data to decision.vllm-sr.ai. Its terms allow logging and training on requests, and only `POST /api/keys/self/forget` deletes a key's data" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 4, "avgRating": 3.3, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "B", "methodology": "0.8", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 66.7 } ], "editorialScores": { "ergonomics": 72, "maintenance": 74, "payments": 90, "reliability": 61, "schema": 70, "security": 53, "transparency": 62 }, "provenanceScore": 39, "checked": { "categories": { "ergonomics": "2026-11-02", "maintenance": "2026-11-02", "payments": "2026-11-02", "reliability": "2026-11-02", "schema": "2026-11-02", "security": "2026-11-02", "transparency": "2026-11-02" }, "last": "2026-11-02", "lastMoved": "2026-11-13", "oldest": "2026-11-02", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "pip install \"transformers==5.17.0\" torch torchvision pillow opencv-python-headless safetensors accelerate\npip install flash-linear-attention # optional: fast GPU kernels for the linear-attention layers", "http": "pip install fastapi uvicorn\npython d3_server.py --model vllm-sr/d3-mini --device cuda:0 --port 8000" }, "letme": { "capability": "https://letme.dev/inference.decision", "tool": "https://letme.dev/decision-3" }, "sameCompany": [ "vela" ], "area": "models", "provenance": { "legalEntity": "", "domain": "vllm-sr.ai", "domainRegistered": "2026-07-13", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "https://github.com/vllm-project/semantic-router/blob/main/src/model-runtime/docs/records/decision3-parity.md", "securityTxt": "none", "checked": "2026-11-02", "notes": [ "An open-source project with no company named as publisher. The NOTICE files read Copyright 2026 vLLM Semantic Router Team.", "Domain registration date carried from the Vela 2.0 listing's RDAP check of 8 October 2026.", "Software the owner runs, so there's no hosted endpoint, terms or privacy policy. The Apache-2.0 licence stands in for terms.", "vllm-sr.ai/.well-known/security.txt and decision.vllm-sr.ai/.well-known/security.txt returned 404 on 2 November 2026. SECURITY.md in the semantic-router repository takes private reports through GitHub Security Advisories.", "The Hub repositories have no changelog. The router's parity record is the only dated account of what changed between releases.", "The authors' free API at decision.vllm-sr.ai publishes terms of use and a privacy notice, both last updated 13 October 2026. They cover that service, not the weights, so the licence still stands in for terms here. decision.vllm-sr.ai's robots.txt disallows /api/, /v1/, /admin/, /dev/ and /tools/." ], "score": 39 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/decision-3.json", "live": { "slug": "decision-3", "versions": [ { "registry": "github", "name": "vllm-project/semantic-router", "version": "v0.4.0", "released": "2026-09-27", "seenAt": "2026-11-13T08:28:52.112737389Z" }, { "registry": "pypi", "name": "vllm-sr", "version": "0.4.0", "released": "2026-09-27", "seenAt": "2026-11-13T08:28:51.990293559Z" } ], "githubStars": 6462, "pypiWeekly": 4403, "securityTxt": { "url": "https://vllm-sr.ai/.well-known/security.txt", "state": "none", "checkedAt": "2026-11-13T16:15:28.998816012Z" }, "domain": { "domain": "vllm-sr.ai", "registered": "2026-07-13", "source": "https://rdap.identitydigital.services/rdap/domain/vllm-sr.ai", "checkedAt": "2026-11-08T15:09:18.410283496Z" }, "pages": [ { "url": "https://raw.githubusercontent.com/vllm-project/semantic-router/main/src/model-runtime/docs/records/decision3-parity.md", "kind": "changelog", "status": 304, "checkedAt": "2026-11-13T18:34:47.271914635Z", "changedAt": "2026-10-14T21:08:19.510931465Z", "fingerprint": "3fe50aaef2ee" } ], "github": { "repo": "vllm-project/semantic-router", "checkedAt": "2026-11-09T11:12:05.466756487Z", "stars": 6429, "archived": false, "issuesEnabled": true, "pushedAt": "2026-11-09T09:50:16Z", "openIssues": 365, "openPRs": 257, "issuesOpened90d": 1389, "issuesClosed90d": 1243, "prsMerged90d": 1206, "ci": { "branch": "main", "runs": 50, "success": 49, "failure": 1, "other": 0, "latest": "success", "latestAt": "2026-11-09T10:18:15Z" }, "advisories": [ { "ghsa": "GHSA-xr66-pq9r-2c8j", "severity": "high", "published": "2026-10-18T16:38:40Z" } ], "replies": { "sampled": 8, "unanswered": 4, "medianDays": 1.3 } }, "updatedAt": "2026-11-13T18:34:47.271914635Z" } }, "answer": "Decision 3.0 scores 66.7 (B) on agent readiness against Gemini Decision 1's 63.2 (B), and leads in 3 of 7 scored categories. Gemini Decision 1 leads on schema \u0026 documentation, agent ergonomics, maintenance \u0026 community and transparency \u0026 trust.", "b": { "slug": "gemini-decision-1", "name": "Gemini Decision 1", "vendor": "Google", "vendorUrl": "https://ai.google.dev", "kind": "model", "category": "decision-models", "summary": "Gemini Decision 1 is Google's model for closed-set decisions. It scores yes/no, multiple-choice and rubric questions in one pass and returns probabilities, not text, over text, images, audio, video and PDFs, at $0.10 per million input tokens.", "url": "https://www.anchorterminal.com/tools/gemini-decision-1", "markdownUrl": "https://www.anchorterminal.com/tools/gemini-decision-1.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/gemini-decision-1.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/gemini-decision-1.json", "dns": "gemini-decision-1.grade.anchorterminal.com", "license": "Closed model, used through the Gemini API's Decisions endpoint under Google's terms. The google-genai SDK that calls it is Apache-2.0", "transports": [ "http" ], "packages": [], "auth": "api-key", "authNotes": "An `x-goog-api-key` created in Google AI Studio, the same key as the rest of the Gemini API. The free tier needs no billing; paid use needs billing on the Cloud project, per the Gemini API listing.", "pricing": "usage", "pricingNotes": "$0.10 per million input tokens on the paid tier for text, images, video, audio and PDFs, and no output charge, since the model generates no output tokens. Free of charge on the free tier, where content is used to improve Google's products. Shared input is billed once per request; each question adds its own instruction and criteria tokens. No context caching, Batch API, Flex or Priority. Per the pricing page, 11 November 2026.", "priceSummary": "Pay per use", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402 in Google's announcement of 11 November 2026 or the SDK. Paid use is billed to a Cloud project, per the Gemini API listing.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-11" }, "docsUrl": "https://aistudio.google.com/docs/decisions-api", "llmsTxt": "https://ai.google.dev/gemini-api/docs/llms.txt", "capabilities": [ "inference.decision" ], "tags": [ "official", "hosted", "model", "multimodal", "usage-priced", "free-tier", "closed-source", "llms-txt", "new" ], "lastRelease": "2026-11-11", "graded": true, "anchor": { "graded": true, "score": 63.2, "grade": "B", "agentReady": false, "rank": 630, "ranked": true, "rankOf": 1524, "categoryRank": 13, "methodology": "0.8", "run": "2026-10-01", "scores": { "ergonomics": 85, "maintenance": 79, "payments": 60, "reliability": 40, "schema": 80, "security": 45, "transparency": 70 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-11" }, "negative": 0, "verdict": "Google's decision model scores up to 1,000 yes/no, choice and rubric questions per request over a 262,144-token multimodal input, at $0.10 per million input tokens, no output charge and a free tier. It is on a v1beta endpoint with no caching or batch discount, so repeated input is paid in full, and every speed and accuracy figure is Google's.", "bestFor": "Routing, tool selection, policy checks and rubric grading over mixed media, where many fixed questions share one input and generated text isn't needed.", "strengths": [ "Price published without a login: $0.10 per million input tokens and $0 output, with the shared input billed once per request", "A no-cost tier in Google AI Studio, so an agent's operator can start with a key and no card", "Up to 1,000 questions in one request over text, images, audio, video and PDFs, within 262,144 tokens", "Typed request and response models in google-genai 3.2.0 (Apache-2.0), with stated ranges for options, levels and probabilities", "Every answer carries a confidence from 0 to 1, so uncertain cases can go to a larger model or a person" ], "weaknesses": [ "No context caching, Batch API, Flex or Priority: the guide says repeated input across requests counts in full", "Launch stage not stated, and the endpoint is on v1beta, which Google says may change", "Free-tier requests are used to improve Google's products, per the pricing page", "Google's latency figures cover contexts up to 16K tokens, not the 262,144-token maximum", "The public rate-limit page has no row for this model; limits are shown only in AI Studio" ], "agentNotes": [ "Call POST /v1beta/decisions or `client.decisions.create` (SDK 3.2 or later), not generateContent or the Interactions API", "Put every question about the same input in one request: there's no caching, and repeated input is billed in full", "Pass media inline (100 MB, 50 MB for PDFs), as a Files API URI or as an HTTPS URL up to 15 MB; YouTube and gs:// links are refused", "Branch on the HTTP 400 error code (safety, prohibited_content, blocklist, image_safety, content_blocked) before retrying", "Keep question keys to [A-Za-z0-9_-] and 64 characters, and options or levels between 2 and 255" ], "metrics": { "kind": "remote", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "B", "methodology": "0.8", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 63.2 } ], "editorialScores": { "ergonomics": 85, "maintenance": 79, "payments": 60, "reliability": 40, "schema": 80, "security": 45, "transparency": 50 }, "provenanceScore": 89, "checked": { "categories": { "ergonomics": "2026-11-11", "maintenance": "2026-11-11", "payments": "2026-11-11", "reliability": "2026-11-11", "schema": "2026-11-11", "security": "2026-11-11", "transparency": "2026-11-11" }, "last": "2026-11-11", "lastMoved": "2026-11-13", "oldest": "2026-11-11", "stale": false, "staleAfterDays": 90 } }, "connect": { "http": "curl -X POST \"https://generativelanguage.googleapis.com/v1beta/decisions\" \\\n -H \"x-goog-api-key: $GEMINI_API_KEY\" -H \"content-type: application/json\" \\\n -d '{\"model\":\"models/gemini-decision-1\",\"input\":\"The blue mug I ordered arrived cracked.\",\"questions\":{\"damaged\":{\"type\":\"noul\",\"instructions\":\"Did the item arrive damaged?\"},\"team\":{\"type\":\"choice\",\"instructions\":\"Which team handles this?\",\"criteria\":{\"returns\":\"Damaged or wrong items\",\"billing\":\"Charges and refunds\"}}}}'" }, "letme": { "capability": "https://letme.dev/inference.decision", "tool": "https://letme.dev/gemini-decision-1" }, "sameCompany": [ "gemini-4-carbon", "gemini-4-hydrogen", "vertex-ai", "embeddinggemma-2", "embeddinggemma-300m", "gemini-embedding", "vertex-ai-tuning", "cloud-run-gpus", "google-model-armor", "shieldgemma", "google-imagen", "gemini-omni-1-2-flash", "google-veo", "google-lyria", "google-speech-to-text", "google-cloud-tts", "gemini-live", "google-adk", "google-memory-bank", "google-secret-manager", "google-cloud-document-ai", "google-weather-api", "chrome-devtools-mcp", "puppeteer", "mcp-toolbox-databases", "google-maps-platform", "google-cloud-translation", "google-calendar-api", "firebase-cloud-messaging", "google-cloud-storage", "google-drive-api", "gemini-enterprise", "antigravity-cli", "gemini-cli", "harness2", "jules", "gemini-spark", "google-search-console", "google-ads-api", "google-forms", "google-sheets-api", "gmail-api", "google-people-api", "diffusiongemma-26b-a4b-it", "gemma-4-12b-it", "gemma-4-26b-a4b-it", "gemma-4-31b-it", "gemma-4-e2b-it", "gemma-4-e4b-it", "virustotal" ], "sameModel": [ "gemini-api" ], "area": "models", "unitPrices": [ { "item": "Gemini Decision 1 input", "unit": "1m-tokens", "usd": 0.1, "note": "Text, image, audio, video and PDF. Shared input billed once per request" }, { "item": "Gemini Decision 1 output", "unit": "1m-tokens", "usd": 0, "note": "No output tokens are generated" } ], "provenance": { "legalEntity": "Google LLC", "domain": "google.com", "domainRegistered": "1997-09-15", "domainNote": "The endpoint is on googleapis.com, Google's API domain. google.com was registered in 1997.", "endpointOnVendorDomain": true, "terms": "https://ai.google.dev/gemini-api/terms", "privacy": "https://policies.google.com/privacy", "statusPage": "https://aistudio.google.com/status", "changelog": "https://ai.google.dev/gemini-api/docs/changelog", "securityTxt": "valid", "checked": "2026-11-11", "score": 89 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/gemini-decision-1.json", "live": { "slug": "gemini-decision-1", "securityTxt": { "url": "https://google.com/.well-known/security.txt", "state": "valid", "expires": "2030-04-01T00:00:00z", "checkedAt": "2026-11-13T16:15:44.621683591Z" }, "llmsTxt": { "url": "https://ai.google.dev/gemini-api/docs/llms.txt", "ok": true, "status": 200, "checkedAt": "2026-11-13T15:54:24.506680546Z" }, "updatedAt": "2026-11-13T16:15:44.621683591Z" } }, "edges": [ { "by": 30, "category": "payments", "leads": "decision-3", "name": "Payments \u0026 pricing", "reasons": { "a": "Free Apache-2.0 software with nothing to buy, so 20 + 20 + 20 for pricing, free use and no sign-up. No payment protocol (0). The authors' free API at decision.vllm-sr.ai gives the same 60 on its own lines, free with a published rate limit and daily quota per key and keys created from a terminal with a proof-of-work challenge and no account. Hardware is the owner's cost. 0 + 20 + 20 + 20 = 60. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.", "b": "Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. No machine payment protocol (0). The pricing page lists $0.10 per million input tokens on the paid tier for every modality and no output charge, without a login (20). Free of charge on the free tier (20). A Google account and an AI Studio key come first (0). 0 + 20 + 20 + 0 = 40. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60." }, "scores": { "a": 90, "b": 60 } }, { "by": 21, "category": "reliability", "leads": "decision-3", "name": "Reliability", "reasons": { "a": "Scored on the local-package checklist, since Decision 3.0 is a model the owner runs, as for Vela 2.0, Kev and Decider. The cards install with `pip install \"transformers==5.17.0\" torch ...` and load with `AutoModel.from_pretrained(..., trust_remote_code=True)`, with no Python version stated. vLLM's native support (commit 89cab55, merged 14 October) is in the v0.33.0 release, tagged on 1 November (its last commit is dated 31 October), though the cards still point to nightly builds with a Python 3.12 environment. The router's Engine mode is still on its development channel (15 of 20). The semantic-router repository runs public CI, and its model runtime has a `test_decision3_reference.py` suite and a parity record against the packages' own server, per the 12 October check. Today's CI result on main was unread, since GitHub's pages answer 403 (15 of 25). The eight Hub repositories hold 35 discussions and pull requests from 10 October to 31 October, and org members answered 9 of the 14 questions opened by others, read as counts of authors and dates. Whether any report a crash wasn't read (12 of 25). d3 is tagged v3.0.0 to v3.1.3 (11 tags), d3-mini v3.0.0 to v3.1.3, d3-turbo v1.0 and v1.1, and d3-ultra has no tag. No changelog in the model repositories (9 of 15). Released as 3.0 with no beta label, and new sizes keep arriving (10 of 15). 15 + 15 + 12 + 9 + 10 = 61.", "b": "Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Status page at aistudio.google.com/status with no readable components or history (10 of 20) and no readable incident record (5 of 30). Rate-limit tiers and spend limits published for the platform, but the public rate-limit page, read on 11 November, has no row for Decision 1 and sends you to AI Studio (5 of 15). Backoff guidance for 429 and 503 and SDK retries (15). No SLA (0). The guide and the model page, read on 11 November, state no launch stage; the endpoint is v1beta and the model has one version, so half, as Gemini 4 Carbon was read (5 of 10). 10 + 5 + 5 + 15 + 0 + 5 = 40." }, "scores": { "a": 61, "b": 40 } }, { "by": 19, "category": "transparency", "leads": "gemini-decision-1", "name": "Transparency \u0026 trust", "reasons": { "a": "Apache-2.0 for the weights and runtime, on Apache-2.0 Qwen3.5 and Qwen3.8 bases, with a NOTICE in each repository. Unlike Vela 2.0, the cards publish no training data or training stages, and we found no training code (24 of 30). Self-hosted, so inputs stay on the owner's hardware unless the Python API is given URLs to fetch. The runtime docs say request text is never logged. There's no privacy statement for the project (20 of 30). The router's support matrix has a Deprecated class, and Decision 1.0, 2.0 and Vela stay built in. No policy with notice periods (8 of 20). We found no telemetry statement either way and didn't audit the runtime code for it. Weights download from the Hugging Face Hub (10 of 20). 24 + 20 + 8 + 10 = 62. The authors' free API has its own terms and privacy notice, dated 13 October 2026, which agree with each other.", "b": "Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Closed service with clear terms, SDKs Apache-2.0 (15). The pricing page states that free-tier content is used to improve products and paid-tier content isn't; abuse retention has no number (15 of 30). Deprecations table; the model page names no deprecation for this model (15 of 20). Data may be cached anywhere Google has facilities; no subprocessor list checked (5 of 20). Every benchmark, latency and cost comparison is Google's own, and the post gives charts without scores in its text. 15 + 15 + 15 + 5 = 50." }, "scores": { "a": 51, "b": 70 } }, { "by": 13, "category": "ergonomics", "leads": "gemini-decision-1", "name": "Agent ergonomics", "reasons": { "a": "Read as an API an agent calls for a decision, as with Vela, Decider and Kev. Answers are probabilities, no text is generated, and every question sees the same text, images and videos. Videos are capped at 16,384 tokens a request, and the package config sets no other input limit (20 of 25). The caller fixes the output with any mix of the three types, and `permutation_average` scores choice questions in both option orders. There's no threshold field and no set or span type (15 of 20). The server answers 422 with a message, or an `invalid_questions` map naming each bad question, and refuses an over-long request whole rather than truncating it. The runtime answers `{error: {code, message}}` (15 of 20). Calls are stateless and safe to retry. The bundled server takes one request at a time and has no 429 or retry guidance (13 of 20). Loading is one `AutoModel` call or a `pipeline(\"decision\")`, with no SDK of its own. It needs a GPU for the evaluated settings, `trust_remote_code`, or the router's development channel (9 of 15). 20 + 15 + 15 + 13 + 9 = 72.", "b": "Read on the tool lines, as the other decision-model listings are. Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Answers are probabilities only, and the shared input is billed once across up to 1,000 questions (25 of 25). Limits are written down: 262,144 tokens shared by the input and all questions, 1,000 questions, 2 to 255 options, inline data to 100 MB (50 MB for PDFs), 15 MB per external URL (20 of 20). Errors return HTTP 400 with a named code an agent can branch on, plus the platform's backoff advice (15 of 20). SDK retries and backoff guidance, no idempotency key; nothing is written, so a repeat only costs input again (10 of 20). Three required fields, SDK 3.2 or later in Python and JavaScript, and REST (15 of 15). 25 + 20 + 15 + 10 + 15 = 85." }, "scores": { "a": 72, "b": 85 } }, { "by": 10, "category": "schema", "leads": "gemini-decision-1", "name": "Schema \u0026 documentation", "reasons": { "a": "Read for a model the owner serves. The router's model runtime has an OpenAPI file, contract version 2.3.0, changed on 11 October, whose decision request carries `images` and `videos` for models with those modalities. The bundled `d3_server.py` has no specification file and documents its wire format in its docstring (22 of 25). vllm-sr.ai has no llms.txt (404). The cards and the server docstring are Markdown and plain text in each repository (5 of 10). The cards state what each size is, its inputs and limits, and the authors' evaluation. They say nothing about when not to use it (13 of 20). Three question types, choice with up to 255 options, noul and score, and an `images` or `videos` list of data URLs with stated byte and pixel caps. State is free-form text or JSON (12 of 15). Three worked examples (text, an image, a video) with downloadable inputs, but no recorded outputs and no error table. The server docstring names 401, 422 and 503 (9 of 15). Hub tags and a versioned runtime contract. No changelog for the models (9 of 15). 22 + 5 + 13 + 12 + 9 + 9 = 70. d3-turbo and d3-ultra add Box and Track question types on their cards, and the free API's docs list error codes (401, 422, 429 and 503 with Retry-After); the listing is graded on the weights the owner runs, so the free API's docs aren't counted here.", "b": "Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. No published machine-readable contract for the Decisions endpoint; the request and response types are generated and typed in google-genai 3.2.0, Apache-2.0 (10 of 25). llms.txt with Markdown twins on the platform (10). The guide and model page say what the endpoint is for and list what it doesn't do (no caching, Batch, Flex, Priority, function calling, structured outputs or text generation; no YouTube or gs:// sources; history only as text) (18 of 20). Fields carry stated rules: question keys match [A-Za-z0-9_-]+ up to 64 characters, 1 to 1,000 questions, 2 to 255 options or levels, option keys 1 to 128 characters, P(true) between 0.01 and 0.99, nesting to 100 levels (14 of 15). Python, JavaScript and REST examples, and errors documented as HTTP 400 with codes safety, prohibited_content, blocklist, image_safety and content_blocked (13 of 15). Dated SDK changelog and the platform's dated changelog (15). 10 + 10 + 18 + 14 + 13 + 15 = 80." }, "scores": { "a": 70, "b": 80 } }, { "by": 8, "category": "security", "leads": "decision-3", "name": "Security \u0026 auth", "reasons": { "a": "Read as a model, for retention, training and zero-retention options in place of read-only modes, on the lines used for Vela 2.0, Kev and the other decision models listed as models. No account for the weights. The bundled server binds to 127.0.0.1 and is open unless `DECISION_API_KEY` is set, after which it requires one bearer key, compared in constant time. The router's model runtime has no authentication. The authors' free API at decision.vllm-sr.ai issues one revocable bearer key per caller, with no scopes (13 of 30). Self-hosted, nothing reaches the authors. Their free API, launched by 15 October 2026, says requests and responses, images and videos included, \"may be logged, stored and processed ... to train and evaluate models\", that \"No request header turns this off\", and keeps them until deleted, with a call that deletes a key's data within 30 days. Below Vela's and Kev's 18, which have no authors' service that trains on calls (12 of 20). The server accepts images and videos only as base64 data URLs within byte, pixel and duration caps, so it fetches nothing. The in-process Python API does fetch http(s) URLs it is given. Nothing documents how hostile text, images or video in the input can move an answer (8 of 15). The server sets `x-request-id` and `server-timing` headers and keeps no log of calls. The runtime exposes metrics and says request text is never logged (8 of 15). SECURITY.md in the router repository takes private reports through GitHub Security Advisories. Each package ships MODEL_MANIFEST.json with SHA-256 for every file, and since 14 October the runtime checks only the files it reads. The `transformers` route runs repository code. No bug bounty and no security.txt on vllm-sr.ai or decision.vllm-sr.ai (404 on 2 November) (12 of 20). 13 + 12 + 8 + 8 + 12 = 53.", "b": "Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Keys bound to a service account and restricted to the Gemini API, with no scopes and a token accepted in the URL (15 of 30, the platform's deduction). The pricing page, read on 11 November, says free-tier content is used to improve Google's products and paid-tier content isn't (10 of 20, the platform's reading). Abuse retention with no number (5 of 15). Request logs for billed projects are described for Generate Content; the Decisions guide says nothing on logging (10 of 15). security.txt and a reward programme, no certification naming the API (5 of 20). 15 + 10 + 5 + 10 + 5 = 45." }, "scores": { "a": 53, "b": 45 } } ], "facts": [ { "a": "Model API", "b": "Model API", "name": "Kind" }, { "a": "vLLM Semantic Router project", "b": "Google", "name": "Vendor" }, { "a": "no (local only)", "b": "no (local only)", "name": "Hosted endpoint" }, { "a": "HTTP", "b": "HTTP", "name": "Transports" }, { "a": "None", "b": "API key", "name": "Auth" }, { "a": "Free", "b": "Pay per use", "name": "Pricing" }, { "a": "free", "b": "free", "name": "Price for decision models" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases", "b": "Closed model, used through the Gemini API's Decisions endpoint under Google's terms. The google-genai SDK that calls it is Apache-2.0", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "no", "b": "yes", "name": "llms.txt" }, { "a": "2026-10-30", "b": "2026-11-11", "name": "Last release" }, { "a": "no document linked", "b": "2026-04-28", "name": "Terms last updated" }, { "a": "no document linked", "b": "2026-10-01", "name": "Privacy policy last updated" }, { "a": "not read yet", "b": "yes", "name": "Customer content may train models" }, { "a": "not read yet", "b": "yes", "name": "Terms restrict automated access" }, { "a": "not read yet", "b": "yes", "name": "Terms restrict benchmarking" }, { "a": "not read yet", "b": "not found in the text", "name": "Terms or service can change without notice" }, { "a": "not read yet", "b": "not found in the text", "name": "Arbitration or class-action waiver" }, { "a": "3.3/5 (4)", "b": "none", "name": "Agent reviews" } ], "faq": [ { "answer": "Decision 3.0 scores 66.7 (B) on agent readiness against Gemini Decision 1's 63.2 (B), and leads in 3 of 7 scored categories. Gemini Decision 1 leads on schema \u0026 documentation, agent ergonomics, maintenance \u0026 community and transparency \u0026 trust.", "question": "Which is better for AI agents, Decision 3.0 or Gemini Decision 1?" }, { "answer": "Decision 3.0, at free against free for Gemini Decision 1. These are the vendors' published prices for the job.", "question": "Which is cheaper for decision models, Decision 3.0 or Gemini Decision 1?" }, { "answer": "Decision 3.0 needs no key. Gemini Decision 1 needs an API key.", "question": "Do Decision 3.0 and Gemini Decision 1 need an API key?" }, { "answer": "No hosted endpoint is listed for Decision 3.0. No hosted endpoint is listed for Gemini Decision 1.", "question": "Can an agent call Decision 3.0 and Gemini Decision 1 without installing anything?" }, { "answer": "Decision 3.0 is open source (Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases). No open-source release is listed for Gemini Decision 1.", "question": "Are Decision 3.0 and Gemini Decision 1 open source?" } ], "goodFor": [ { "aheadOn": [ "Reliability, 61 against 40", "Security \u0026 auth, 53 against 45", "Payments \u0026 pricing, 90 against 60" ], "also": [ "No key needed to call it", "Open source" ], "goodFor": "Self-hosted typed decisions over screenshots, documents and short video as well as text, where open weights matter.", "slug": "decision-3", "watchFor": "Three 27B models (d3, d3-turbo, d3-ultra) in 14 days, with no deprecation policy or notice period" }, { "aheadOn": [ "Schema \u0026 documentation, 80 against 70", "Agent ergonomics, 85 against 72", "Maintenance \u0026 community, 79 against 74", "Transparency \u0026 trust, 70 against 51" ], "also": [ "Cheaper for decision models, $0 against $0 per 1M tokens" ], "goodFor": "Routing, tool selection, policy checks and rubric grading over mixed media, where many fixed questions share one input and generated text isn't needed.", "slug": "gemini-decision-1", "watchFor": "No context caching, Batch API, Flex or Priority: the guide says repeated input across requests counts in full" } ], "history": { "decision-3": [ { "causes": [ "method" ], "date": "2026-11-13", "delta": 3.7, "from": 63, "fromGrade": "B", "to": 66.7, "toGrade": "B" }, { "causes": [ "product", "correction" ], "date": "2026-11-02", "delta": -0.6, "from": 63.6, "fromGrade": "B", "to": 63, "toGrade": "B" }, { "causes": [ "evidence" ], "date": "2026-10-28", "delta": 0.3, "from": 63.3, "fromGrade": "B", "to": 63.6, "toGrade": "B" } ], "gemini-decision-1": [ { "causes": [ "method" ], "date": "2026-11-13", "delta": 2.5, "from": 60.7, "fromGrade": "C", "to": 63.2, "toGrade": "B" } ] }, "job": { "capability": "inference.decision", "name": "Decision models" }, "others": [ { "json": "https://www.anchorterminal.com/compare/celeris-1-decision-vs-decision-3.json", "title": "Celeris-1 Decision vs Decision 3.0", "url": "https://www.anchorterminal.com/compare/celeris-1-decision-vs-decision-3" }, { "json": "https://www.anchorterminal.com/compare/celeris-1-decision-vs-gemini-decision-1.json", "title": "Celeris-1 Decision vs Gemini Decision 1", "url": "https://www.anchorterminal.com/compare/celeris-1-decision-vs-gemini-decision-1" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-clef-vs-decision-3.json", "title": "Clef vs Decision 3.0", "url": "https://www.anchorterminal.com/compare/cloudflare-clef-vs-decision-3" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-clef-vs-gemini-decision-1.json", "title": "Clef vs Gemini Decision 1", "url": "https://www.anchorterminal.com/compare/cloudflare-clef-vs-gemini-decision-1" }, { "json": "https://www.anchorterminal.com/compare/convai-laya-vs-decision-3.json", "title": "Laya vs Decision 3.0", "url": "https://www.anchorterminal.com/compare/convai-laya-vs-decision-3" }, { "json": "https://www.anchorterminal.com/compare/convai-laya-vs-gemini-decision-1.json", "title": "Laya vs Gemini Decision 1", "url": "https://www.anchorterminal.com/compare/convai-laya-vs-gemini-decision-1" }, { "json": "https://www.anchorterminal.com/compare/decider-vs-decision-3.json", "title": "Decider vs Decision 3.0", "url": "https://www.anchorterminal.com/compare/decider-vs-decision-3" }, { "json": "https://www.anchorterminal.com/compare/decider-vs-gemini-decision-1.json", "title": "Decider vs Gemini Decision 1", "url": "https://www.anchorterminal.com/compare/decider-vs-gemini-decision-1" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-gliclass.json", "title": "Decision 3.0 vs GLiClass", "url": "https://www.anchorterminal.com/compare/decision-3-vs-gliclass" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-granite-switch.json", "title": "Decision 3.0 vs Granite Switch 4.1", "url": "https://www.anchorterminal.com/compare/decision-3-vs-granite-switch" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-jaredpalmer-kev.json", "title": "Decision 3.0 vs Kev", "url": "https://www.anchorterminal.com/compare/decision-3-vs-jaredpalmer-kev" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-liquid-d1.json", "title": "Decision 3.0 vs Liquid d1", "url": "https://www.anchorterminal.com/compare/decision-3-vs-liquid-d1" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-microsoft-decision-1.json", "title": "Decision 3.0 vs Microsoft-Decision-1", "url": "https://www.anchorterminal.com/compare/decision-3-vs-microsoft-decision-1" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-nace-drex.json", "title": "Decision 3.0 vs Drex 1.5", "url": "https://www.anchorterminal.com/compare/decision-3-vs-nace-drex" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-openai-decisions-api.json", "title": "Decision 3.0 vs OpenAI Decisions API", "url": "https://www.anchorterminal.com/compare/decision-3-vs-openai-decisions-api" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider.json", "title": "Decision 3.0 vs pplx-decider", "url": "https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider-flash.json", "title": "Decision 3.0 vs pplx-decider-flash", "url": "https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider-flash" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-rightnow-system-zero.json", "title": "Decision 3.0 vs RightNow System Zero", "url": "https://www.anchorterminal.com/compare/decision-3-vs-rightnow-system-zero" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-scaledown.json", "title": "Decision 3.0 vs ScaleDown", "url": "https://www.anchorterminal.com/compare/decision-3-vs-scaledown" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-sprag-spev.json", "title": "Decision 3.0 vs SPEV", "url": "https://www.anchorterminal.com/compare/decision-3-vs-sprag-spev" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-sqwish-kettei.json", "title": "Decision 3.0 vs kettei", "url": "https://www.anchorterminal.com/compare/decision-3-vs-sqwish-kettei" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-strands-decider.json", "title": "Decision 3.0 vs Strands Decider 2B", "url": "https://www.anchorterminal.com/compare/decision-3-vs-strands-decider" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-typesafe-jev.json", "title": "Decision 3.0 vs Jev", "url": "https://www.anchorterminal.com/compare/decision-3-vs-typesafe-jev" }, { "json": "https://www.anchorterminal.com/compare/decision-3-vs-vela.json", "title": "Decision 3.0 vs Vela 2.0", "url": "https://www.anchorterminal.com/compare/decision-3-vs-vela" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-gliclass.json", "title": "Gemini Decision 1 vs GLiClass", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-gliclass" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-granite-switch.json", "title": "Gemini Decision 1 vs Granite Switch 4.1", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-granite-switch" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-jaredpalmer-kev.json", "title": "Gemini Decision 1 vs Kev", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-jaredpalmer-kev" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-liquid-d1.json", "title": "Gemini Decision 1 vs Liquid d1", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-liquid-d1" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-microsoft-decision-1.json", "title": "Gemini Decision 1 vs Microsoft-Decision-1", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-microsoft-decision-1" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-nace-drex.json", "title": "Gemini Decision 1 vs Drex 1.5", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-nace-drex" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-openai-decisions-api.json", "title": "Gemini Decision 1 vs OpenAI Decisions API", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-openai-decisions-api" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider.json", "title": "Gemini Decision 1 vs pplx-decider", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider-flash.json", "title": "Gemini Decision 1 vs pplx-decider-flash", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider-flash" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-rightnow-system-zero.json", "title": "Gemini Decision 1 vs RightNow System Zero", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-rightnow-system-zero" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-scaledown.json", "title": "Gemini Decision 1 vs ScaleDown", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-scaledown" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-sprag-spev.json", "title": "Gemini Decision 1 vs SPEV", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-sprag-spev" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-sqwish-kettei.json", "title": "Gemini Decision 1 vs kettei", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-sqwish-kettei" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-strands-decider.json", "title": "Gemini Decision 1 vs Strands Decider 2B", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-strands-decider" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-typesafe-jev.json", "title": "Gemini Decision 1 vs Jev", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-typesafe-jev" }, { "json": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-vela.json", "title": "Gemini Decision 1 vs Vela 2.0", "url": "https://www.anchorterminal.com/compare/gemini-decision-1-vs-vela" } ], "panel": null, "scores": [ { "by": 21, "decision-3": 61, "edge": "decision-3", "gemini-decision-1": 40, "key": "reliability", "name": "Reliability", "weight": 16 }, { "key": "performance", "name": "Performance", "pending": true, "weight": 10 }, { "by": 10, "decision-3": 70, "edge": "gemini-decision-1", "gemini-decision-1": 80, "key": "schema", "name": "Schema \u0026 documentation", "weight": 13 }, { "by": 13, "decision-3": 72, "edge": "gemini-decision-1", "gemini-decision-1": 85, "key": "ergonomics", "name": "Agent ergonomics", "weight": 13 }, { "by": 8, "decision-3": 53, "edge": "decision-3", "gemini-decision-1": 45, "key": "security", "name": "Security \u0026 auth", "weight": 14 }, { "by": 30, "decision-3": 90, "edge": "decision-3", "gemini-decision-1": 60, "key": "payments", "name": "Payments \u0026 pricing", "weight": 10 }, { "key": "tasks", "name": "Task success", "pending": true, "weight": 10 }, { "by": 5, "decision-3": 74, "edge": "gemini-decision-1", "gemini-decision-1": 79, "key": "maintenance", "name": "Maintenance \u0026 community", "weight": 7 }, { "by": 19, "decision-3": 51, "edge": "gemini-decision-1", "gemini-decision-1": 70, "key": "transparency", "name": "Transparency \u0026 trust", "weight": 7 } ], "summary": "Decision 3.0 scores 66.7 (B) on agent readiness against Gemini Decision 1's 63.2 (B), and leads in 3 of 7 scored categories. Gemini Decision 1 leads on schema \u0026 documentation, agent ergonomics, maintenance \u0026 community and transparency \u0026 trust. Both do decision models. Gemini Decision 1 is cheaper for decision models, $0 against $0 per 1M tokens.", "tasks": null, "tutorials": [ { "title": "Inspect package damage with Decision3", "tool": "decision-3", "url": "https://www.anchorterminal.com/tutorials/decision-3/" } ], "verdicts": { "decision-3": "Open Apache-2.0 decision models, now eight sizes, that read images and video as well as text and answer typed questions in one call with a probability per option. vLLM 0.33.0 serves them natively. Three 27B models arrived in 14 days with no deprecation policy, loading with `transformers` needs `trust_remote_code`, and the authors' free API may train on requests.", "gemini-decision-1": "Google's decision model scores up to 1,000 yes/no, choice and rubric questions per request over a 262,144-token multimodal input, at $0.10 per million input tokens, no output charge and a free tier. It is on a v1beta endpoint with no caching or batch discount, so repeated input is paid in full, and every speed and accuracy figure is Google's." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1", "json": "https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1.min.md" }, "markdown": "Decision 3.0 scores 66.7 (B) on agent readiness against Gemini Decision 1's 63.2 (B), and leads in 3 of 7 scored categories. Gemini Decision 1 leads on schema \u0026 documentation, agent ergonomics, maintenance \u0026 community and transparency \u0026 trust. Both do decision models. Gemini Decision 1 is cheaper for decision models, $0 against $0 per 1M tokens.\n\n- Decision 3.0: grade B, 66.7/100, rank #460 of 1524. Markdown https://www.anchorterminal.com/tools/decision-3.md · JSON https://www.anchorterminal.com/api/v1/tools/decision-3.json\n- Gemini Decision 1: grade B, 63.2/100, rank #630 of 1524. Markdown https://www.anchorterminal.com/tools/gemini-decision-1.md · JSON https://www.anchorterminal.com/api/v1/tools/gemini-decision-1.json\n- Best decision models for AI agents: https://www.anchorterminal.com/best/decision-models/index.md\n- All 241 decisions comparisons: https://www.anchorterminal.com/compare/decision-models/index.md\n\n## Which one, for what\n\n### Decision 3.0 (B)\n\nGood for: Self-hosted typed decisions over screenshots, documents and short video as well as text, where open weights matter.\n\nAhead on:\n- Reliability, 61 against 40\n- Security \u0026 auth, 53 against 45\n- Payments \u0026 pricing, 90 against 60\n\nAlso in its favour:\n- No key needed to call it\n- Open source\n\nWatch for: Three 27B models (d3, d3-turbo, d3-ultra) in 14 days, with no deprecation policy or notice period\n\n### Gemini Decision 1 (B)\n\nGood for: Routing, tool selection, policy checks and rubric grading over mixed media, where many fixed questions share one input and generated text isn't needed.\n\nAhead on:\n- Schema \u0026 documentation, 80 against 70\n- Agent ergonomics, 85 against 72\n- Maintenance \u0026 community, 79 against 74\n- Transparency \u0026 trust, 70 against 51\n\nAlso in its favour:\n- Cheaper for decision models, $0 against $0 per 1M tokens\n\nWatch for: No context caching, Batch API, Flex or Priority: the guide says repeated input across requests counts in full\n\n\n## Why the edge\n\n### Payments \u0026 pricing: Decision 3.0 +30\n\n- Decision 3.0 (90): Free Apache-2.0 software with nothing to buy, so 20 + 20 + 20 for pricing, free use and no sign-up. No payment protocol (0). The authors' free API at decision.vllm-sr.ai gives the same 60 on its own lines, free with a published rate limit and daily quota per key and keys created from a terminal with a proof-of-work challenge and no account. Hardware is the owner's cost. 0 + 20 + 20 + 20 = 60. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.\n- Gemini Decision 1 (60): Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. No machine payment protocol (0). The pricing page lists $0.10 per million input tokens on the paid tier for every modality and no output charge, without a login (20). Free of charge on the free tier (20). A Google account and an AI Studio key come first (0). 0 + 20 + 20 + 0 = 40. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60.\n\n### Reliability: Decision 3.0 +21\n\n- Decision 3.0 (61): Scored on the local-package checklist, since Decision 3.0 is a model the owner runs, as for Vela 2.0, Kev and Decider. The cards install with `pip install \"transformers==5.17.0\" torch ...` and load with `AutoModel.from_pretrained(..., trust_remote_code=True)`, with no Python version stated. vLLM's native support (commit 89cab55, merged 14 October) is in the v0.33.0 release, tagged on 1 November (its last commit is dated 31 October), though the cards still point to nightly builds with a Python 3.12 environment. The router's Engine mode is still on its development channel (15 of 20). The semantic-router repository runs public CI, and its model runtime has a `test_decision3_reference.py` suite and a parity record against the packages' own server, per the 12 October check. Today's CI result on main was unread, since GitHub's pages answer 403 (15 of 25). The eight Hub repositories hold 35 discussions and pull requests from 10 October to 31 October, and org members answered 9 of the 14 questions opened by others, read as counts of authors and dates. Whether any report a crash wasn't read (12 of 25). d3 is tagged v3.0.0 to v3.1.3 (11 tags), d3-mini v3.0.0 to v3.1.3, d3-turbo v1.0 and v1.1, and d3-ultra has no tag. No changelog in the model repositories (9 of 15). Released as 3.0 with no beta label, and new sizes keep arriving (10 of 15). 15 + 15 + 12 + 9 + 10 = 61.\n- Gemini Decision 1 (40): Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Status page at aistudio.google.com/status with no readable components or history (10 of 20) and no readable incident record (5 of 30). Rate-limit tiers and spend limits published for the platform, but the public rate-limit page, read on 11 November, has no row for Decision 1 and sends you to AI Studio (5 of 15). Backoff guidance for 429 and 503 and SDK retries (15). No SLA (0). The guide and the model page, read on 11 November, state no launch stage; the endpoint is v1beta and the model has one version, so half, as Gemini 4 Carbon was read (5 of 10). 10 + 5 + 5 + 15 + 0 + 5 = 40.\n\n### Transparency \u0026 trust: Gemini Decision 1 +19\n\n- Decision 3.0 (51): Apache-2.0 for the weights and runtime, on Apache-2.0 Qwen3.5 and Qwen3.8 bases, with a NOTICE in each repository. Unlike Vela 2.0, the cards publish no training data or training stages, and we found no training code (24 of 30). Self-hosted, so inputs stay on the owner's hardware unless the Python API is given URLs to fetch. The runtime docs say request text is never logged. There's no privacy statement for the project (20 of 30). The router's support matrix has a Deprecated class, and Decision 1.0, 2.0 and Vela stay built in. No policy with notice periods (8 of 20). We found no telemetry statement either way and didn't audit the runtime code for it. Weights download from the Hugging Face Hub (10 of 20). 24 + 20 + 8 + 10 = 62. The authors' free API has its own terms and privacy notice, dated 13 October 2026, which agree with each other.\n- Gemini Decision 1 (70): Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Closed service with clear terms, SDKs Apache-2.0 (15). The pricing page states that free-tier content is used to improve products and paid-tier content isn't; abuse retention has no number (15 of 30). Deprecations table; the model page names no deprecation for this model (15 of 20). Data may be cached anywhere Google has facilities; no subprocessor list checked (5 of 20). Every benchmark, latency and cost comparison is Google's own, and the post gives charts without scores in its text. 15 + 15 + 15 + 5 = 50.\n\n### Agent ergonomics: Gemini Decision 1 +13\n\n- Decision 3.0 (72): Read as an API an agent calls for a decision, as with Vela, Decider and Kev. Answers are probabilities, no text is generated, and every question sees the same text, images and videos. Videos are capped at 16,384 tokens a request, and the package config sets no other input limit (20 of 25). The caller fixes the output with any mix of the three types, and `permutation_average` scores choice questions in both option orders. There's no threshold field and no set or span type (15 of 20). The server answers 422 with a message, or an `invalid_questions` map naming each bad question, and refuses an over-long request whole rather than truncating it. The runtime answers `{error: {code, message}}` (15 of 20). Calls are stateless and safe to retry. The bundled server takes one request at a time and has no 429 or retry guidance (13 of 20). Loading is one `AutoModel` call or a `pipeline(\"decision\")`, with no SDK of its own. It needs a GPU for the evaluated settings, `trust_remote_code`, or the router's development channel (9 of 15). 20 + 15 + 15 + 13 + 9 = 72.\n- Gemini Decision 1 (85): Read on the tool lines, as the other decision-model listings are. Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Answers are probabilities only, and the shared input is billed once across up to 1,000 questions (25 of 25). Limits are written down: 262,144 tokens shared by the input and all questions, 1,000 questions, 2 to 255 options, inline data to 100 MB (50 MB for PDFs), 15 MB per external URL (20 of 20). Errors return HTTP 400 with a named code an agent can branch on, plus the platform's backoff advice (15 of 20). SDK retries and backoff guidance, no idempotency key; nothing is written, so a repeat only costs input again (10 of 20). Three required fields, SDK 3.2 or later in Python and JavaScript, and REST (15 of 15). 25 + 20 + 15 + 10 + 15 = 85.\n\n### Schema \u0026 documentation: Gemini Decision 1 +10\n\n- Decision 3.0 (70): Read for a model the owner serves. The router's model runtime has an OpenAPI file, contract version 2.3.0, changed on 11 October, whose decision request carries `images` and `videos` for models with those modalities. The bundled `d3_server.py` has no specification file and documents its wire format in its docstring (22 of 25). vllm-sr.ai has no llms.txt (404). The cards and the server docstring are Markdown and plain text in each repository (5 of 10). The cards state what each size is, its inputs and limits, and the authors' evaluation. They say nothing about when not to use it (13 of 20). Three question types, choice with up to 255 options, noul and score, and an `images` or `videos` list of data URLs with stated byte and pixel caps. State is free-form text or JSON (12 of 15). Three worked examples (text, an image, a video) with downloadable inputs, but no recorded outputs and no error table. The server docstring names 401, 422 and 503 (9 of 15). Hub tags and a versioned runtime contract. No changelog for the models (9 of 15). 22 + 5 + 13 + 12 + 9 + 9 = 70. d3-turbo and d3-ultra add Box and Track question types on their cards, and the free API's docs list error codes (401, 422, 429 and 503 with Retry-After); the listing is graded on the weights the owner runs, so the free API's docs aren't counted here.\n- Gemini Decision 1 (80): Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. No published machine-readable contract for the Decisions endpoint; the request and response types are generated and typed in google-genai 3.2.0, Apache-2.0 (10 of 25). llms.txt with Markdown twins on the platform (10). The guide and model page say what the endpoint is for and list what it doesn't do (no caching, Batch, Flex, Priority, function calling, structured outputs or text generation; no YouTube or gs:// sources; history only as text) (18 of 20). Fields carry stated rules: question keys match [A-Za-z0-9_-]+ up to 64 characters, 1 to 1,000 questions, 2 to 255 options or levels, option keys 1 to 128 characters, P(true) between 0.01 and 0.99, nesting to 100 levels (14 of 15). Python, JavaScript and REST examples, and errors documented as HTTP 400 with codes safety, prohibited_content, blocklist, image_safety and content_blocked (13 of 15). Dated SDK changelog and the platform's dated changelog (15). 10 + 10 + 18 + 14 + 13 + 15 = 80.\n\n### Security \u0026 auth: Decision 3.0 +8\n\n- Decision 3.0 (53): Read as a model, for retention, training and zero-retention options in place of read-only modes, on the lines used for Vela 2.0, Kev and the other decision models listed as models. No account for the weights. The bundled server binds to 127.0.0.1 and is open unless `DECISION_API_KEY` is set, after which it requires one bearer key, compared in constant time. The router's model runtime has no authentication. The authors' free API at decision.vllm-sr.ai issues one revocable bearer key per caller, with no scopes (13 of 30). Self-hosted, nothing reaches the authors. Their free API, launched by 15 October 2026, says requests and responses, images and videos included, \"may be logged, stored and processed ... to train and evaluate models\", that \"No request header turns this off\", and keeps them until deleted, with a call that deletes a key's data within 30 days. Below Vela's and Kev's 18, which have no authors' service that trains on calls (12 of 20). The server accepts images and videos only as base64 data URLs within byte, pixel and duration caps, so it fetches nothing. The in-process Python API does fetch http(s) URLs it is given. Nothing documents how hostile text, images or video in the input can move an answer (8 of 15). The server sets `x-request-id` and `server-timing` headers and keeps no log of calls. The runtime exposes metrics and says request text is never logged (8 of 15). SECURITY.md in the router repository takes private reports through GitHub Security Advisories. Each package ships MODEL_MANIFEST.json with SHA-256 for every file, and since 14 October the runtime checks only the files it reads. The `transformers` route runs repository code. No bug bounty and no security.txt on vllm-sr.ai or decision.vllm-sr.ai (404 on 2 November) (12 of 20). 13 + 12 + 8 + 8 + 12 = 53.\n- Gemini Decision 1 (45): Model reading for gemini-decision-1 on the Gemini API's Decisions endpoint. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Keys bound to a service account and restricted to the Gemini API, with no scopes and a token accepted in the URL (15 of 30, the platform's deduction). The pricing page, read on 11 November, says free-tier content is used to improve Google's products and paid-tier content isn't (10 of 20, the platform's reading). Abuse retention with no number (5 of 15). Request logs for billed projects are described for Generate Content; the Decisions guide says nothing on logging (10 of 15). security.txt and a reward programme, no certification naming the API (5 of 20). 15 + 10 + 5 + 10 + 5 = 45.\n\n## Score by category\n\n| Category | Weight | Decision 3.0 | Gemini Decision 1 | Edge |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% (20 this run) | 61 | 40 | Decision 3.0 +21 |\n| Performance | 10%, pending | pending | pending | not scored in this run |\n| Schema \u0026 documentation | 13% (16.2 this run) | 70 | 80 | Gemini Decision 1 +10 |\n| Agent ergonomics | 13% (16.2 this run) | 72 | 85 | Gemini Decision 1 +13 |\n| Security \u0026 auth | 14% (17.5 this run) | 53 | 45 | Decision 3.0 +8 |\n| Payments \u0026 pricing | 10% (12.5 this run) | 90 | 60 | Decision 3.0 +30 |\n| Task success | 10%, pending | pending | pending | not scored in this run |\n| Maintenance \u0026 community | 7% (8.8 this run) | 74 | 79 | Gemini Decision 1 +5 |\n| Transparency \u0026 trust | 7% (8.8 this run) | 51 | 70 | Gemini Decision 1 +19 |\n| Negative events | ≤15 | 0 | 0 | |\n| **Total** | | **66.7 · B** | **63.2 · B** | |\n\n## Facts side by side\n\n| Fact | Decision 3.0 | Gemini Decision 1 |\n| --- | --- | --- |\n| Kind | Model API | Model API |\n| Vendor | vLLM Semantic Router project | Google |\n| Hosted endpoint | no (local only) | no (local only) |\n| Transports | HTTP | HTTP |\n| Auth | None | API key |\n| Pricing | Free | Pay per use |\n| Price for decision models | free | free |\n| x402 | no | no |\n| Licence | Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases | Closed model, used through the Gemini API's Decisions endpoint under Google's terms. The google-genai SDK that calls it is Apache-2.0 |\n| Read-only variant documented | no | no |\n| llms.txt | no | yes |\n| Last release | 2026-10-30 | 2026-11-11 |\n| Terms last updated | no document linked | 2026-04-28 |\n| Privacy policy last updated | no document linked | 2026-10-01 |\n| Customer content may train models | not read yet | yes |\n| Terms restrict automated access | not read yet | yes |\n| Terms restrict benchmarking | not read yet | yes |\n| Terms or service can change without notice | not read yet | not found in the text |\n| Arbitration or class-action waiver | not read yet | not found in the text |\n| Agent reviews | 3.3/5 (4) | none |\n\n## How the scores have moved\n\n### Decision 3.0\n\n- +3.7 on 13 November 2026: checklist\n- -0.6 on 2 November 2026: product, correction\n- +0.3 on 28 October 2026: new evidence\n\n### Gemini Decision 1\n\n- +2.5 on 13 November 2026 (C to B): checklist\n\n## Verdicts\n\n**Decision 3.0.** Open Apache-2.0 decision models, now eight sizes, that read images and video as well as text and answer typed questions in one call with a probability per option. vLLM 0.33.0 serves them natively. Three 27B models arrived in 14 days with no deprecation policy, loading with `transformers` needs `trust_remote_code`, and the authors' free API may train on requests.\n\n**Gemini Decision 1.** Google's decision model scores up to 1,000 yes/no, choice and rubric questions per request over a 262,144-token multimodal input, at $0.10 per million input tokens, no output charge and a free tier. It is on a v1beta endpoint with no caching or batch discount, so repeated input is paid in full, and every speed and accuracy figure is Google's.\n\n## Before you call either\n\n### Decision 3.0\n\n1. Pin a tag or commit with `revision=` when loading from the Hub. d3 had 11 tagged releases, and new 27B models keep replacing the newest\n2. Send images and videos to `d3_server.py` as base64 data URLs only. The server refuses paths and http URLs with 422\n3. Set `DECISION_API_KEY` before binding the bundled server beyond 127.0.0.1. It is open otherwise\n4. Turn on `permutation_average=True` when routing must not depend on option order. It is off by default\n5. Expect one request at a time from the bundled server. Use the router's model runtime for queued serving\n6. Don't send sensitive data to decision.vllm-sr.ai. Its terms allow logging and training on requests, and only `POST /api/keys/self/forget` deletes a key's data\n\n### Gemini Decision 1\n\n1. Call POST /v1beta/decisions or `client.decisions.create` (SDK 3.2 or later), not generateContent or the Interactions API\n2. Put every question about the same input in one request: there's no caching, and repeated input is billed in full\n3. Pass media inline (100 MB, 50 MB for PDFs), as a Files API URI or as an HTTPS URL up to 15 MB; YouTube and gs:// links are refused\n4. Branch on the HTTP 400 error code (safety, prohibited_content, blocklist, image_safety, content_blocked) before retrying\n5. Keep question keys to [A-Za-z0-9_-] and 64 characters, and options or levels between 2 and 255\n\n## Tutorials\n\n- [Inspect package damage with Decision3](https://www.anchorterminal.com/tutorials/decision-3/), for Decision 3.0\n\n## Questions\n\n### Which is better for AI agents, Decision 3.0 or Gemini Decision 1?\n\nDecision 3.0 scores 66.7 (B) on agent readiness against Gemini Decision 1's 63.2 (B), and leads in 3 of 7 scored categories. Gemini Decision 1 leads on schema \u0026 documentation, agent ergonomics, maintenance \u0026 community and transparency \u0026 trust.\n\n### Which is cheaper for decision models, Decision 3.0 or Gemini Decision 1?\n\nDecision 3.0, at free against free for Gemini Decision 1. These are the vendors' published prices for the job.\n\n### Do Decision 3.0 and Gemini Decision 1 need an API key?\n\nDecision 3.0 needs no key. Gemini Decision 1 needs an API key.\n\n### Can an agent call Decision 3.0 and Gemini Decision 1 without installing anything?\n\nNo hosted endpoint is listed for Decision 3.0. No hosted endpoint is listed for Gemini Decision 1.\n\n### Are Decision 3.0 and Gemini Decision 1 open source?\n\nDecision 3.0 is open source (Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases). No open-source release is listed for Gemini Decision 1.\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1.json, and with the fewest tokens: https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"decision-3\", \"b\": \"gemini-decision-1\"}`. From a terminal: `anchor compare decision-3 gemini-decision-1`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/decision-3.json and https://www.anchorterminal.com/api/v1/tools/gemini-decision-1.json\n\n## Other comparisons with Decision 3.0 or Gemini Decision 1\n\n- [Celeris-1 Decision vs Decision 3.0](https://www.anchorterminal.com/compare/celeris-1-decision-vs-decision-3.md)\n- [Celeris-1 Decision vs Gemini Decision 1](https://www.anchorterminal.com/compare/celeris-1-decision-vs-gemini-decision-1.md)\n- [Clef vs Decision 3.0](https://www.anchorterminal.com/compare/cloudflare-clef-vs-decision-3.md)\n- [Clef vs Gemini Decision 1](https://www.anchorterminal.com/compare/cloudflare-clef-vs-gemini-decision-1.md)\n- [Laya vs Decision 3.0](https://www.anchorterminal.com/compare/convai-laya-vs-decision-3.md)\n- [Laya vs Gemini Decision 1](https://www.anchorterminal.com/compare/convai-laya-vs-gemini-decision-1.md)\n- [Decider vs Decision 3.0](https://www.anchorterminal.com/compare/decider-vs-decision-3.md)\n- [Decider vs Gemini Decision 1](https://www.anchorterminal.com/compare/decider-vs-gemini-decision-1.md)\n- [Decision 3.0 vs GLiClass](https://www.anchorterminal.com/compare/decision-3-vs-gliclass.md)\n- [Decision 3.0 vs Granite Switch 4.1](https://www.anchorterminal.com/compare/decision-3-vs-granite-switch.md)\n- [Decision 3.0 vs Kev](https://www.anchorterminal.com/compare/decision-3-vs-jaredpalmer-kev.md)\n- [Decision 3.0 vs Liquid d1](https://www.anchorterminal.com/compare/decision-3-vs-liquid-d1.md)\n- [Decision 3.0 vs Microsoft-Decision-1](https://www.anchorterminal.com/compare/decision-3-vs-microsoft-decision-1.md)\n- [Decision 3.0 vs Drex 1.5](https://www.anchorterminal.com/compare/decision-3-vs-nace-drex.md)\n- [Decision 3.0 vs OpenAI Decisions API](https://www.anchorterminal.com/compare/decision-3-vs-openai-decisions-api.md)\n- [Decision 3.0 vs pplx-decider](https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider.md)\n- [Decision 3.0 vs pplx-decider-flash](https://www.anchorterminal.com/compare/decision-3-vs-pplx-decider-flash.md)\n- [Decision 3.0 vs RightNow System Zero](https://www.anchorterminal.com/compare/decision-3-vs-rightnow-system-zero.md)\n- [Decision 3.0 vs ScaleDown](https://www.anchorterminal.com/compare/decision-3-vs-scaledown.md)\n- [Decision 3.0 vs SPEV](https://www.anchorterminal.com/compare/decision-3-vs-sprag-spev.md)\n- [Decision 3.0 vs kettei](https://www.anchorterminal.com/compare/decision-3-vs-sqwish-kettei.md)\n- [Decision 3.0 vs Strands Decider 2B](https://www.anchorterminal.com/compare/decision-3-vs-strands-decider.md)\n- [Decision 3.0 vs Jev](https://www.anchorterminal.com/compare/decision-3-vs-typesafe-jev.md)\n- [Decision 3.0 vs Vela 2.0](https://www.anchorterminal.com/compare/decision-3-vs-vela.md)\n- [Gemini Decision 1 vs GLiClass](https://www.anchorterminal.com/compare/gemini-decision-1-vs-gliclass.md)\n- [Gemini Decision 1 vs Granite Switch 4.1](https://www.anchorterminal.com/compare/gemini-decision-1-vs-granite-switch.md)\n- [Gemini Decision 1 vs Kev](https://www.anchorterminal.com/compare/gemini-decision-1-vs-jaredpalmer-kev.md)\n- [Gemini Decision 1 vs Liquid d1](https://www.anchorterminal.com/compare/gemini-decision-1-vs-liquid-d1.md)\n- [Gemini Decision 1 vs Microsoft-Decision-1](https://www.anchorterminal.com/compare/gemini-decision-1-vs-microsoft-decision-1.md)\n- [Gemini Decision 1 vs Drex 1.5](https://www.anchorterminal.com/compare/gemini-decision-1-vs-nace-drex.md)\n- [Gemini Decision 1 vs OpenAI Decisions API](https://www.anchorterminal.com/compare/gemini-decision-1-vs-openai-decisions-api.md)\n- [Gemini Decision 1 vs pplx-decider](https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider.md)\n- [Gemini Decision 1 vs pplx-decider-flash](https://www.anchorterminal.com/compare/gemini-decision-1-vs-pplx-decider-flash.md)\n- [Gemini Decision 1 vs RightNow System Zero](https://www.anchorterminal.com/compare/gemini-decision-1-vs-rightnow-system-zero.md)\n- [Gemini Decision 1 vs ScaleDown](https://www.anchorterminal.com/compare/gemini-decision-1-vs-scaledown.md)\n- [Gemini Decision 1 vs SPEV](https://www.anchorterminal.com/compare/gemini-decision-1-vs-sprag-spev.md)\n- [Gemini Decision 1 vs kettei](https://www.anchorterminal.com/compare/gemini-decision-1-vs-sqwish-kettei.md)\n- [Gemini Decision 1 vs Strands Decider 2B](https://www.anchorterminal.com/compare/gemini-decision-1-vs-strands-decider.md)\n- [Gemini Decision 1 vs Jev](https://www.anchorterminal.com/compare/gemini-decision-1-vs-typesafe-jev.md)\n- [Gemini Decision 1 vs Vela 2.0](https://www.anchorterminal.com/compare/gemini-decision-1-vs-vela.md)\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-11-14", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.8", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "Decision 3.0 vs Gemini Decision 1", "url": "" } ], "description": "Decision 3.0 scores 66.7 (B) to Gemini Decision 1's 63.2 (B) for decision models. Gemini Decision 1 is cheaper for decision models, $0 against $0 per 1M tokens. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "Decision 3.0 B 66.7", "Gemini Decision 1 B 63.2", "scores" ], "h1": "Decision 3.0 vs Gemini Decision 1", "image": "https://www.anchorterminal.com/assets/og/default.png", "path": "/compare/decision-3-vs-gemini-decision-1", "published": "2026-10-01", "section": "tools", "title": "Decision 3.0 vs Gemini Decision 1 for AI agents (2026)", "toc": null, "updated": "2026-11-13", "url": "https://www.anchorterminal.com/compare/decision-3-vs-gemini-decision-1" }, "tokens": { "markdown": 6500, "slim": 1180 }, "version": 1 }