Decision 3.0 Free API
⚠️ Why not a model
Hosted API serves the existing Decision 3.0 checkpoints; it publishes no separate decision-trained weights of its own.
Technical specs
- Base LLMs
- vLLM-SR d3 family (six sizes, 0.6B–27B)
- Decision types
- choice, score, noul
- Features
- hosted-api, systemone-compatible, free-tier, api-key, text, image, video, box-questions, status-page
- License
- Proprietary hosted service; served model checkpoints are Apache-2.0
Links
The Decision 3.0 Free API serves the six existing Decision 3.0 models through the vLLM Semantic Router’s hosted API. The launch post on 14 October 2026 described free access without an account and compatibility with SystemOne.
Access and contract
The current API docs clarify the access flow: inference requests require an API key, and a user can create that key without an account. POST /v1/systemone (also POST /v1/decisions) accepts typed questions; GET /v1/systemone/models lists the served models and their supported question types. The public /api/info endpoint lists limits and rate limits, and /api/status reports service state and incidents.
The service lists six checkpoints from 0.6B to 27B. Text, image and video input are supported at model-specific limits. The two larger models list box in addition to choice, noul and score; smaller models list the latter three. The API is free at the time of review; check the live docs for current limits.
Why this is not a model
This is a hosted inference runtime over the Decision 3.0 checkpoints already catalogued on the site. The API does not publish an additional checkpoint or decision-trained weights of its own, so it belongs in Runtimes and stays off the quadrant.
Vendor claims
The launch post says Decision 3.0 is “#1 on both Jev Decision Index” and serves six models from 0.6B to 27B. The current API documentation confirms the six-model lineup and request formats; the ranking claim is not established by the API docs and remains UnverifiedClaim. See the model card for the board evidence and methodology.
Board check, 17 Oct 2026. > publicado (OK do Fabricio 17/10 BRT). On the board’s 16 Oct data the 27B scores Full 66.26 (#2, behind StartLux-Decision v1.1 27B at 69.12) and public 65.45 (#5), so the “#1” in the launch post held for the 12-13 Oct board but not for the latest one. The site header now shows “#1 Text 66.26, #1 Vision 71.26”; we did not verify the vision figure. See Decision 3.0.