Category · Models & inference
Decision models for AI agents
Models that answer typed questions with probabilities instead of text, a yes or no, one label from a set or a level on a rubric, for routing, triage and checks inside agents and workflows. Compared on accuracy, calibration, context, latency, price and whether the weights are open.
Best decision models for AI agents, the shortlist263 head-to-head comparisons
Capability keys inference.decision · All tools
letme.dev/inference.decision picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 48 | OpenAI Decisions APIOpenAI · Model API | Decisions | A | 78.5 | 3.5 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
The Decisions API is an OpenAI endpoint, generally available since 22 October 2026, that reads text, images and JSON and returns typed answers with probabilities, a predicate, a choice from a fixed set or a rubric score. Top strength Three typed question kinds, predicate, choice and score, with per-option probabilities and up to 400 questions about one input in a call Top weakness One model alias, |
||||||||||||
| 168 | Drex 1.5Nace AI, Inc. (Nace.AI) · Model API | Decisions | BB | 73.6 | 3.5 (4) | n/a | n/a | Freemium | API key | Hosted | ||
|
Drex 1.5 is Nace.AI's decision model. It returns a probability for each option in typed yes or no, choice and score questions. Hosted at console.nace.ai, with open weights under the Nace AI Open RAIL-M licence. Top strength Public OpenAPI 3.1 document with POST /v1/systemone, /v1/decisions and GET /v1/models, plus llms.txt and a reference for limits and errors. Top weakness drex-v1.0 was retired on 21 October 2026, 27 days after release, and its requests now go to Drex Max at Max's limits and price, with no notice period published. |
||||||||||||
| 205 | DeciderMark Marosi (Mapika) · Model API | Decisions | BB | 72.8 | 3.3 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Decider is a family of open-weight decision models by Mark Marosi (Mapika), from 0.8B to 35B parameters under Apache-2.0. It answers typed yes or no, choice and score questions with probabilities, and runs locally from the `decider-ai` Python package. Top strength Apache-2.0 code and weights, with the training code, data builders and per-version measurements in the repository Top weakness The local server has no authentication option. It binds to 127.0.0.1 since 1.7.1 |
||||||||||||
| 221 | LayaConvai Innovations · Model API | Decisions | BB | 72.4 | 2.5 (2) | n/a | n/a | Free · OSS | None | Local | ||
|
Open-source decision engine from Convai Innovations, released under Apache-2.0. Top strength Apache-2.0 code and weights, installed with Top weakness Base checkpoints score 0.362 and 0.352 on the maintainers' typed-decisions benchmark against a 0.318 random baseline, so it needs fine-tuning |
||||||||||||
| 249 | pplx-deciderPerplexity · Model API | Decisions | BB | 71.8 | 3.5 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
pplx-decider is Perplexity's 27B decision model, a fine-tune of Qwen3.8-27B. It answers yes or no, choice and score questions about text, JSON and images with probabilities, through Perplexity's hosted Decisions API or open weights on Hugging Face. Top strength Weights for v1.2 are public on Hugging Face under Apache-2.0, with the Top weakness Requests naming |
||||||||||||
| 293 | KevJared Palmer · Model API | Decisions | BB | 70.5 | 3.5 (2) | n/a | n/a | Free · OSS | None | Local | ||
|
Kev is a family of four open-weight decision models by Jared Palmer, released together as Kev 1.0 on 1 October 2026 under Apache-2.0. Top strength Apache-2.0 code, adapters and heads on Apache-2.0 Qwen bases, with release tarballs and SHA-256 checksums for the 0.8B, 4B and 9B models Top weakness No package. |
||||||||||||
| 295 | Vela 2.0vLLM Semantic Router project (with KR Labs) · Model API | Decisions | BB | 70.4 | 4.3 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Vela 2.0 is a family of four open-weight decision models from the vLLM Semantic Router project and KR Labs, released on 6 October 2026 under Apache-2.0 for routing, safety checks, personal-data spans and hallucination checks. Top strength Five question types in one request (choice, noul, score, set and span), with span answers as labelled character offsets and a probability each Top weakness No version tags on the four Hub repositories, and the 4B and 9B weights were replaced in place on 3 October 2026 |
||||||||||||
| 341 | RightNow System ZeroRightNow AI, Inc. · Model API | Decisions | B | 69.3 | none | n/a | n/a | $0.0001 / tx | API key | Hosted | ||
|
RightNow System Zero is a hosted, closed-weight decision model from RightNow AI, Inc. It answers boolean, choice and score questions about text, JSON, images, WAV audio and PDFs with per-option probabilities, over HTTP or a WebSocket stream. Top strength Boolean, choice and score questions, up to 1,024 a call, answered with a probability per option and no output charge Top weakness Standard context is 38,000 tokens. The 64k, 128k and 1M bands are held, and video is a preview |
||||||||||||
| 375 | pplx-decider-flashPerplexity · Model API | Decisions | B | 68.4 | 3.3 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
pplx-decider-flash is Perplexity's 4B decision model, a Qwen3.5-4B fine-tune distilled from the 27B pplx-decider. It answers yes or no, choice and score questions about text, JSON and images with probabilities, through the hosted Decisions API or open weights. Top strength $0.01 per million input tokens, half the 27B model's price, with free output and no per-request fee Top weakness Released on 16 October 2026, with a 31-day record for the endpoint and no Decisions component on the status page |
||||||||||||
| 380 | ClefCloudflare · Model API | Decisions | B | 68.3 | 3.0 (2) | n/a | n/a | Freemium | API key | Hosted | ||
|
Clef, Clef-flash and Clef-omni are Cloudflare's open-weight decision models, hosted on Workers AI with Apache-2.0 weights. Clef-omni, released on 9 October 2026, also takes audio and video. Top strength Apache-2.0 weights for Clef, Clef-flash and Clef-omni on Hugging Face, ungated, with no account needed to download Top weakness Hosted Clef-flash's context fell from 64K to 24K tokens on 9 October 2026, announced in the changelog the day it took effect |
||||||||||||
| 461 | Decision 3.0vLLM Semantic Router project · Model API | Decisions | B | 66.7 | 3.3 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
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. Top strength Text, images and video in one request, with every question seeing all of them Top weakness Three 27B models (d3, d3-turbo, d3-ultra) in 14 days, with no deprecation policy or notice period |
||||||||||||
| 555 | Strands Decider 2BAmazon Web Services (Strands Agents) · Model API | Decisions | B | 64.8 | 2.0 (1) | n/a | n/a | Free · OSS | None | Local | ||
|
Strands Decider 2B is an open-weight decision model from AWS's Strands Labs, released on 1 October 2026 under Apache-2.0. It answers typed yes or no, choice and score questions with probabilities, and runs locally from a Python package. Top strength Apache-2.0 code and weights, with the training recipe, data inventory and evaluation logs published Top weakness Package version 0.x, described as experimental in its metadata, and CHANGELOG.md lists model releases but no package versions |
||||||||||||
| 632 | Gemini Decision 1Google · Model API | Decisions | B | 63.2 | none | n/a | n/a | Pay per use | API key | Local | ||
|
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. Top strength Price published without a login: $0.10 per million input tokens and $0 output, with the shared input billed once per request Top weakness No context caching, Batch API, Flex or Priority: the guide says repeated input across requests counts in full |
||||||||||||
| 662 | Granite Switch 4.1IBM · Model API | Decisions | B | 62.5 | 3.0 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Granite Switch 4.1 is IBM's open-weight 8B and 30B Granite 4.1 checkpoint with a decision adapter, a safety adapter and a speech encoder, served by the owner as one OpenAI-compatible vLLM endpoint for chat, decisions and transcription. Top strength Weights are ungated on Hugging Face under Apache 2.0, in 8B and 30B, with the composer code under the same licence Top weakness The decision adapter has no typed request format. The caller writes the prompt strings and computes single-token label ids with the tokeniser |
||||||||||||
| 684 | JevTypeSafe AI · Model API | Decisions | B | 62.1 | 3.0 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Jev is TypeSafe AI's first System One model, a closed decision model behind an HTTP API. Top strength Typed answers with probabilities for Top weakness Early access behind a waitlist, with no free tier or free credits found |
||||||||||||
| 722 | ketteiSqwish Labs · Model API | Decisions | C | 61.2 | 3.8 (4) | n/a | n/a | $0.50 / tx | API key | Hosted | ||
|
kettei is Sqwish Labs' family of decision models, released on 14 October 2026. Its OneHot API returns a probability for each allowed answer to yes or no, choice and rating questions about text or images, and fine-tunes the smaller sizes. Top strength Binary, single and ordinal decisions, up to 128 a request, each with a probability for every allowed answer and an optional abstain or cost policy Top weakness Launched on 14 October 2026, with no status page, no changelog and no SLA |
||||||||||||
| open | JEV-27B-VLAutoTrust AI Lab · Open-weight model | Decisions | C | 59.5 | 2.5 (4) | n/a | n/a | Free | None | Local | ||
|
JEV-27B-VL is AutoTrust AI Lab's open-weight decision model, a LoRA adapter and decision head on Qwen3.8-27B that returns option probabilities for questions over text and images, trained on labels distilled from TypeSafe's Jev. It is not a TypeSafe product. Top strength Calibrated probabilities for yes or no, up to 256 options or a 0 to 5 rating, in one forward pass Top weakness Trained on labels distilled from TypeSafe's hosted Jev, whose customer agreement forbids distillation |
||||||||||||
| open | joey-1Baseten · Open-weight model | Decisions | C | 57.1 | none | n/a | n/a | Free | None | Local | ||
|
joey-1 is Base Labs' open-weight 2B decision model, a full fine-tune of Qwen3.5-2B that takes text and images and returns a probability for every option of typed questions about a state in one prefill. Base Labs is Baseten's research lab. Top strength Probabilities for choice, yes or no and score questions, several per request from one shared prefill, over text and images Top weakness Serving it needs OneJev at a pinned commit plus the repository's serve.patch, with no released runtime support |
||||||||||||
| open | Drex Max v1Nace AI, Inc. (Nace.AI) · Open-weight model | Decisions | C | 55.5 | none | n/a | n/a | Free | None | Local | ||
|
Drex Max v1 is Nace.AI's open-weight 27B decision model, a fine-tune of Qwen3.8-27B that takes text, JSON and images with typed questions and returns a probability for every option without generating text. Top strength Probabilities for choice, yes or no and score questions over text, JSON and images, with nothing generated Top weakness The licence bars commercial use above $1 million in revenue or funding and by anyone with a competing product, and restricts the outputs too |
||||||||||||
| 962 | GLiClassKnowledgator · Model API | Decisions | C | 55.5 | none | n/a | n/a | Free · OSS | None | Local | ||
|
GLiClass is an open-source Python library and family of open-weight zero-shot text classifiers from Knowledgator. It scores every candidate label in one forward pass and runs locally through a pipeline or a Ray Serve endpoint. Top strength Apache-2.0 code and weights, with Top weakness No calibration figures on the v3.0 cards, which report F1 only; the v4 2B and 27B repositories hold calibration report files we couldn't read |
||||||||||||
| 981 | Liquid d1Liquid AI · Model API | Decisions | C | 54.9 | 3.0 (4) | n/a | n/a | Freemium | API key | Hosted | ||
|
d1 is Liquid AI's decision model family. It answers typed yes or no, choice and score questions about text and images with probabilities, through the hosted d1.1 and d1-ultra models and the open-weight d1-2B, d1-omni-2B and d1-omni-600M. Top strength $0.02 per million input tokens for Top weakness The terms of 30 September 2026 allow Liquid to train on API content, and the data page doesn't say which endpoints carry zero retention |
||||||||||||
| open | elephant-1Baseten · Open-weight model | Decisions | D | 53.6 | 2.3 (4) | n/a | n/a | Free | None | Local | ||
|
elephant-1 is Base Labs' open-weight 9B decision model, a full fine-tune of Qwen3.5-9B that returns a probability for every option of typed questions about a state in one prefill. Base Labs is Baseten's research lab. Top strength Probabilities for choice, yes or no and score questions, several per request from one shared prefill Top weakness Serving it needs OneJev at a pinned commit plus the repository's serve.patch, with no released runtime support |
||||||||||||
| 1115 | Celeris-1 DecisionCeleris (Marqo Inc) · Model API | Decisions | D | 51 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
Celeris-1 Decision is a hosted multimodal diffusion model from Celeris, a Marqo Inc research lab. Its HTTP API answers typed questions about text, structured data and images with probabilities and optional explanations. Top strength Typed binary, choice and score answers, with multiple questions about one state per request Top weakness Early-access terms and capacity-dependent workspace activation |
||||||||||||
| 1119 | SPEVSprag · Model API | Decisions | D | 50.9 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
SPEV is Sprag's open-weight decision model, released on 14 October 2026. It answers predicate, choice and score questions about text, audio, an image or video frames with a probability for each answer, through Sprag's hosted API or Apache-2.0 weights. Top strength Native audio evidence, so turn detection and call triage need no transcription step first Top weakness Published on 14 October 2026, with no service record, no status page and no API changelog |
||||||||||||
| open | mammoth-1Baseten · Open-weight model | Decisions | D | 48.6 | 2.3 (4) | n/a | n/a | Free | None | Local | ||
|
mammoth-1 is Base Labs' open-weight 27B decision model, a full fine-tune of Qwen3.5-27B that returns a probability for every option of each question about a state, without generating text. Base Labs is Baseten's research lab. Top strength Several choice, yes-or-no and score questions answered together from one shared-prefix prefill, with no text to parse Top weakness Runs as trained only on OneJev's qev runtime at a pinned commit with the repository's patch |
||||||||||||
| 1267 | Microsoft-Decision-1.1Microsoft · Model API | Decisions | E | 44.8 | none | n/a | n/a | Pay per use | OAuth or key | Local | ||
|
Microsoft-Decision-1.1 is a Microsoft model, built on Gemma 4 26B A4B, that scores fixed answer options with a calibrated probability for each, from text or images. Generally available on Foundry, OpenRouter and Vercel from $0.042 per million input tokens. Top strength Generally available on Foundry, OpenRouter and Vercel Top weakness No machine-readable contract or reference page with the request schema was found |
||||||||||||
| 1319 | Microsoft-Decision-1Microsoft · Model API | Decisions | E | 42.1 | 2.8 (4) | n/a | n/a | Pay per use | OAuth or key | Local | ||
|
Microsoft-Decision-1 is a Microsoft model, post-trained from Qwen3.5-9B, that scores fixed answer options with a calibrated probability for each. It is in public preview on Microsoft Foundry and OpenRouter, at $0.042 per million input tokens. Top strength Price published without a login: $0.042 per million input tokens, with output free, on the US and EU Datazone deployments and on OpenRouter Top weakness Public preview only. Microsoft's post gives no general availability date and states no licence for the hosted model |
||||||||||||
| 1364 | ScaleDownCarbon ScaleDown, Inc. · Model API | Decisions | E | 39.6 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
ScaleDown's API serves four task-specific models, Classify, Extract, Summarize and Compress, called with an API key at api.scaledown.xyz. Classify answers Jev-shaped decision requests with probabilities. Top strength
Top weakness No status page, no published rate limits and no SLA in the docs |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
How we test this category
A fixed set of labelled decisions (yes or no, one label from a set, a level on a rubric) sent to every model in the same request shape, with the same states and questions. We check accuracy against the labels, calibration (expected calibration error, Brier score and how often an answer given 0.9 or more is wrong), whether answers move when the option order changes, latency and the cost per 1,000 decisions.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.