1. Public benchmarks
Standard datasets run with fixed prompts, versions and decoding settings.
EVALUATION
Measured. Reproducible.
Published with Evidence.
OUTAI will publish results only after the relevant model exists, the test procedure is documented and the evidence can be reproduced.
4Evaluation layers
0Invented scores
100%Claim traceability target
∞Continuous re-evaluation
How we evaluate
2. Internal task suites
Representative enterprise and regional tasks with controlled ground truth.
3. Adversarial testing
Prompt injection, ambiguity, unsupported claims and unsafe edge cases.
4. Human review
Expert review where automated scoring is insufficient.
Publication table
| Model | Status | Scores | Evidence |
|---|---|---|---|
| OUTAI Base | Planned | Not published | Pending model selection |
| OUTAI Reason | Planned | Not published | Pending training |
| OUTAI Code | Planned | Not published | Pending training |
| OUTAI Medical | Research | Not published | Clinical validation required |
This website intentionally contains no fabricated benchmark claims. Verified results can replace these placeholders after formal evaluation.