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.
Version 2.2 (4 October) continued training with more decision data; v1, v1.1 and v2 stay as tags, and the registry's earlier 2.0.0 version holds the weights uploaded on 27 September. In the maker's served runs (SGLang 0.5.20 and jev-adapter, temperature 1.65, 1 to 3 October), v2.2 scores 100% on JevBench's public easy tier, 98.61% standard and 56.76% hard; 82.50% on a 600-item held-out decision set; 82.67% on questions with 53 to 151 options; and 40.00% on long-document questions. Up to 255 options with served wording (labels A to Z, then AA, AB and so on), evaluated up to 151; native wording still caps at 26. The temperature was not refit for v2.2, and calibration was last measured on v2 (hard-tier ECE 0.184). The maker's own run of the third-party Decision Index 0.2.1 gives 41.14. 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 (also tagged v2.1 and v2.2), LoRA repositories, FP8 and GGUF builds. Also served on the Standard API from $0.019 per million input tokens, output free. An earlier version was ranked 45th on JevBench v1.4.2 (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 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 | 56.8% |
| Reported latency | 22.6 ms adapter time for one warm request on one AMD MI350X (v2.2 example); 25.8 ms p50 / 41.9 ms p95 on one H200 measured on v2 |
Hosted API
Served by Standard Thinking — $0/MTok output. Get access · .