Standard One 8B
Standard Thinking's decision model. Ministral 3 8B with a merged LoRA, served through stock SGLang and an open jev-adapter that exposes /v1/systemone and scores the supplied options in one forward pass with per-type temperatures.
Version 2 (26 September) replaced the weights; v1 and v1.1 stay as tags. The maker reports 71.1% on a 400-case typed-decisions suite (2,000 decisions) and, on JevBench's public tiers, 100% easy, 93.1% standard and 55.0% hard (Jev 1.13 in its run: 72.1% on hard), with hard-tier ECE 0.184. At most 26 options per question; trained in nine languages; the Pixtral vision tower is kept but image decisions are not benchmarked. Training data is listed with its licences, several of them CC BY-SA. Siblings: Standard One 3B, LoRA repositories, FP8 and GGUF builds. Also served on the Standard API from $0.019 per million input tokens, output free. JevBench v1.4.2 ranked it 45th (76.6% public, 26.6% sealed).
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one calibrated probability
- classify — assigns a category from a fixed taxonomy
- route — sends the input to one of several destinations
At a glance
| Parameters | 8B |
| Base model | mistralai/ministral-3-8b-instruct-2512-bf16 |
| Maker | Standard Thinking |
| Released | 2026-09-24 |
| License | apache-2.0 |
| Reported accuracy | 71.1% |
| Reported latency | 25.8 ms p50 / 41.9 ms p95 raw serial on one H200, SGLang 0.5.20, ~280 input tokens per decision |
Hosted API
Served by Standard Thinking — $0/MTok output. Get access · API docs.
Get the weights
pip install systemonemodels
systemone pull standard-thinking/standard-one
The files are served from the maker's Hugging Face repository, , and verified against the checksums recorded here.