Gemini Decision 1 vs Vela 2.0

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.

Vela 2.0 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). There is no hosted API. Hardware is the owner's cost. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 60 is 90.

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.

Vela 2.0 63

Read as a model, for retention, training and zero-retention options in place of read-only modes, on the lines used for Kev and the other decision models listed as models. No account. The bundled vela2_serve.py binds to 127.0.0.1 and is open unless VELA2_API_KEY is set, after which it requires one bearer key, compared in constant time. The router's model runtime has no authentication, and its docs say to expose it only on a trusted network (13 of 30). Self-hosted, so nothing reaches the authors and no caller data can train a model. The model runtime reference says request text is never logged and metrics carry no request content. Nothing says what the router's other components keep (18 of 20, level with Kev). The models are trained to flag prompt attacks, each question sees only the state and its own block, and the router's SECURITY.md names adversarial input to classification models in its threat model. Nothing documents how hostile text in the state can move an answer (10 of 15). The runtime exposes Prometheus metrics and, on request, the revision, digest and timings that answered a call. The bundled server keeps no record of calls (9 of 15). SECURITY.md takes private reports through GitHub Security Advisories, with severity classes and a disclosure policy, re-read in the clone on 1 November. Weights are safetensors with a SHA256SUMS manifest in the decoder repositories, and the runtime checks every file's SHA-256 and never runs code from a model repository. The transformers route does run repository code through trust_remote_code=True, with no tag to pin. No bug bounty, and no security.txt on vllm-sr.ai (404 on 1 November) (13 of 20). 13 + 18 + 10 + 9 + 13 = 63.

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.

Vela 2.0 57

Scored on the local-package checklist, since Vela 2.0 is a model the owner runs. There are two routes. The model cards load it with pip install torch "transformers>=5.17" and AutoModel.from_pretrained(..., trust_remote_code=True), with no Python version stated. The router's vllm-sr package on PyPI states Python 3.10 or newer, but serving a model with it is newer than the 0.4.0 stable release and needs the development channel (12 of 20). The semantic-router repository has public CI whose model-runtime suite includes five test_vela2_* files, and each model repository ships a parity record against the research scorer with zero decision differences on 230 rows. We couldn't read today's CI result on main, since the GitHub API was rate-limited and the Actions page we got was stale (15 of 25). The router repository showed 408 open issues on 8 October, 23 of them mentioning Vela, against 200 commits between 23 September and 8 October. The four Hub repositories have no discussions. Issue #4668 of 7 October tracks the 0.3B's CPU latency, which the release note puts at 79 ms against 16 ms for Vela 1.0 (14 of 25). The Hub repositories have no version tags, and the 4B and 9B weights were replaced in place on 3 October. The router pins the revisions it loads and publishes a dated release note (6 of 15). Released as 2.0 with no beta label on the cards, though the vllm-sr package is classified beta and the family is two days past launch (10 of 15). Re-read on 1 November from a clone. No stable release since 0.4.0 (27 September), the vllm-sr package on main is 0.5.0, still classified beta and needing Python 3.10 or newer, and the open issue count and CI results rest on the 8 October check, since GitHub's pages answered 403. 12 + 15 + 14 + 6 + 10 = 57.

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.

Vela 2.0 54

Apache-2.0 for the weights, code and documentation, on Apache-2.0 Decision 2.0 and Qwen3.5 bases, with MODIFICATIONS.md and ATTRIBUTIONS.md in each repository. The 0.3B's tokeniser keeps the Gemma Terms of Use, and training data isn't redistributed and keeps its own licences, some CC BY-SA. We found no training code (26 of 30). Self-hosted, so inputs stay on the owner's hardware. The runtime docs say request text is never logged and metrics carry no request content. There's no privacy statement for the project (22 of 30). The router's support matrix defines a Deprecated class and says nothing is in it, and the release note gives the lines that restore each Vela 1.0 model. No policy with notice periods (10 of 20). We found no telemetry statement either way and didn't audit the runtime's code for it. Weights download from the Hugging Face Hub (10 of 20).

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.

Vela 2.0 79

Read as an API an agent calls for a decision, as with Jev, Clef and Kev. Answers are probabilities and spans, output_tokens is always 0, and the decoders read the state once and fork it per question. Input is 16,384 tokens a sequence, or 8,192 on the 0.3B (20 of 25). The caller fixes the output shape with any mix of the five types, a threshold per question, a head override for spans and /v1/bundle for several tasks in one call (18 of 20). The runtime answers with {error: {code, message}}, and a failed question carries its own code without failing the others. The bundled server answers 422 with loc, msg and type, and 413 for oversize questions (16 of 20). Calls are stateless and safe to retry. The runtime answers 429 overloaded past 256 queued requests a model. We found no Retry-After header in the runtime code and no retry guidance (15 of 20). typesafe-sdk works against the bundled server per USAGE.md, and loading is one AutoModel call. Setup needs trust_remote_code, about 17 GB of GPU memory for the 4B, or the router's development channel (10 of 15).

Gemini Decision 1 79

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. Released on 11 November 2026, the day of this check (30). Deprecations with earliest dates and no stated minimum notice (4 of 12). Shutdowns in the last 90 days kept to their table (4 of 8). Dated changelogs: google-genai 3.2.0's entry adds the Decisions API; the platform changelog wasn't re-read today (11 of 15). SDK replies not sampled (5 of 10). google-genai 3.2.0 tagged the same day with client.decisions.create; the JavaScript SDK is per Google's post (15). Runtimes stated on the platform (10). 30 + 4 + 4 + 11 + 5 + 15 + 10 = 79. The model is hours old, so these lines are the platform's habits more than this model's record.

Vela 2.0 74

Read as a model, for deprecation notice and model churn rather than release counts, on the lines used for Kev, Decider, Laya and the other decision models listed as models. Vela 2.0 launched on 6 October 2026, and the four Hub repositories were last changed on 7 October (card edits only). 26 days on 1 November (30). No deprecation policy or notice period. The 4B and 9B weights were replaced in place on 3 October, the repositories have no tags, and the router pins the revisions it loads. When Vela 3.0's vela3-fast replaced Vela 2.0 0.3B as the router's default on 28 October (pull request #5512), the docs gave the line that restores the 0.3B the same day, and the support matrix still lists nothing as deprecated (5 of 12, level with Kev's and Decider's 5). No Vela 2.0 weights have changed since 3 October and pinned revisions still load, but Vela 3.0, six new models launched on 26 October, took over the router's default 22 days after Vela 2.0's launch (5 of 8, level with Laya's 5 for a default switch with a restore path). The router repository's 200 newest commits span 13 October to 1 November under 57 author names, and the four Hub repositories have no discussions. Issue reply times rest on the 8 October check, since GitHub's pages answered 403 (17 of 25). No SDK of its own. The cards point to typesafe-sdk, and the clone's publish workflow ships the vllm-sr CLI to PyPI on a stable and a development channel, with 0.4.0 the last stable tag. pypi.org didn't resolve from the shell on 1 November, and the 8 October reading of PyPI's JSON endpoint, a path its robots.txt disallows, is struck (10 of 15). The router repository runs CI, a security scan and a package check, with today's result on main unread (7 of 10). 30 + 5 + 5 + 17 + 10 + 7 = 74.