Runs itself
A scheduled workflow refreshes public rankings and evaluates every configured model without manual scoring.
CONTINUOUS · AUTONOMOUS · AUDITABLE
AI evaluation that
shows its work.
A third-generation testFAILS implementation that runs objective cases through configured model APIs, synchronizes official Arena leaderboards, records failures, and publishes provenance without pretending unlike measurements are interchangeable.
Fails visibly
Network errors, invalid JSON, skipped modalities, and assertion failures remain in the evidence record.
Separates evidence
Arena preference data is displayed beside testFAILS3 results, never folded into the autonomous score.
LATEST AUTONOMOUS EVALUATION
Measured by executable assertions.
The stored run belongs to an earlier suite.
Its evidence remains preserved, but it is not displayed as a result for suite 2026.10.2. Run the current 70-probe manifest before making a comparison.
stored 965f2b823344 · current 3646641d3af5OFFICIAL EXTERNAL CONTEXT
Arena leaderboards, synchronized automatically.
Retrieved 2026-10-15T15:06:46Z · CC BY 4.0 dataset ↗
Published 2026-10-13 · text_style_control · top 25 synchronized rows
- 1gemini-4-argon-highgoogle1524.8
- 2muse-wonder-maxmeta1512.7
- 3claude-opus-5.5-highanthropic1508.1
- 4claude-fable-5-highanthropic1503.4
- 5claude-opus-4-6-highanthropic1503.1
- 6claude-opus-4-7-highanthropic1501.0
- 7claude-fable-5.1-maxanthropic1500.6
- 8claude-opus-4-6anthropic1497.7
- 9gemini-3.8-flash-highgoogle1496.3
- 10muse-spark-1.2 (xHigh)meta1494.6
Published 2026-10-14 · agent · top 25 synchronized rows
- 1Claude Opus 5.5 (Max)anthropic0.172
- 2GPT 6.1 Sol (Max)openai0.148
- 3Claude Fable 5.1 (Max)anthropic0.141
- 4Gemini 4 Argon (High)google0.137
- 5GPT 6 Astra (Max)openai0.137
- 6Claude Sonnet 5.5 (Max)anthropic0.125
- 7GPT 6 Sol (Max)openai0.112
- 8Claude Opus 5 (Max)anthropic0.092
- 9Claude Fable 5 (High)anthropic0.084
- 10GPT 5.6 Sol (xHigh)openai0.074
Published 2026-10-13 · search_style_control · top 25 synchronized rows
- 1gpt-5.6-sol-xhighopenai1257.9
- 2gpt-6.1-sol-maxopenai1249.9
- 3gpt-6-astra-maxopenai1246.2
- 4claude-fable-5.1-highanthropic1233.4
- 5qwen3.8-maxalibaba1229.5
- 6claude-fable-5-highanthropic1229.0
- 7gpt-6-sol-maxopenai1228.2
- 8gpt-5.5-searchopenai1226.5
- 9claude-opus-4-6-searchanthropic1222.0
- 10claude-opus-5-highanthropic1214.4
Published 2026-10-13 · webdev · top 25 synchronized rows
- 1claude-opus-5.5-maxanthropic1811.1
- 2gpt-6-astra-maxopenai1780.2
- 3claude-sonnet-5.5-xhighanthropic1773.7
- 4gpt-6.1-sol-maxopenai1760.1
- 5claude-fable-5.1-maxanthropic1742.7
- 6claude-sonnet-5.5-highanthropic1713.0
- 7claude-opus-5-maxanthropic1690.0
- 8gpt-6-sol-maxopenai1685.9
- 9gemini-4-argon-highgoogle1678.0
- 10qwen3.8-max-0902alibaba1673.2
Arena ranks are human-preference or IPS measurements published by Arena. testFAILS3 scores are executable case assertions. They answer different questions and are intentionally not combined.