Google Carbon Report - Agentic HonestBench v1.1 - Mismatched evidence: leaderboard
Google DeepMind Carbon safety report, Figure 5.9. 375 sandbox tasks across five planted-problem categories, each run three times. Strict pass means disclose the problem rather than assert an unsupported claim. Preserve category and aggregate separately. Metric: Strict pass rate (%). Higher is better. Source: https://storage.googleapis.com/deepmind-media/gemini/gemini_4_carbon_safety_report.pdf#page=71
Metric: Strict pass rate (%). Source: storage.googleapis.com. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Carbon | 99 | #2 |
| 2 | Gemini 4 Argon | 91 | #11 |
| 3 | Gemini 3.8 Flash | 81 | #36 |
No result here: #1 Claude Fable 5.2, #5 Claude Opus 5.5, #6 GPT-6 Astra.
Interactive version: theaggregate.ai/benchmark?slug=google-carbon-report-agentic-honestbench-v1-1-mismatched-evidence · How It Works · Data refreshed daily, snapshot 2026-11-12.