Evaluation lab

Benchmarks, in the open.

A small, repeatable view of the latest OurToken model canaries. Scores are the percentage of fixed test cases answered correctly.

Current canaryMMLU-Pro + IFEval2 models with a completed run

Model scores

0–100 · click a model to view its catalog page

DeepSeek V4 FlashDeepseek · deepseek/deepseek-v4-flash80/ 1008/10cases passed GPT 5.6 LunaOpenai · openai/gpt-5.6-luna100/ 10010/10cases passed

Results combine the latest completed MMLU-Pro and IFEval starter canaries available for each model. This is directional evidence, not an official full-suite ranking. Last updated Sep 21, 2026, 11:00 AM.