CAIS Risk Index
Claude Opus 5.5 26.2 leads since Sep 2026 Not in model cards
Averages seven safety measures that correlate weakly with general capability. They cover agent prompt injection, bioweapons help, political bias, overconfidence, lying under pressure and harmful or deceptive play in text games.
Members on 2026-11-01: Agent Red Teaming, the VCT, HPCT and BioTIER refusal average, PCT, HLE miscalibration, MASK, MACHIAVELLI and TextQuests Harm. Lower is better.
Made by Center for AI Safety Released Dec 2025 Composite indices All tasks public
Last model added Oct 2026
Scores refreshed 2 Nov 2026
- Level
- Research frontier Researchers, on open problems or tasks that take days
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
65
Models scored on its own leaderboard
Claude Opus 5.5 leads with 26.2
Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.
- 1 Claude Opus 5.5 high 26.2
- 2 o3 high 30.2
- 3 Claude Fable 5.1 high 30.3
- 4 Claude Fable 5 high 30.8
- 5
Muse Wonder high 32.4 - 6 Muse Spark 1.1 high 32.5
- 7 GPT-6 Astra high 35.5
- 8 o4-mini high 35.9
- 9 Claude Opus 5 high 36.1
- 10 Claude Opus 4.7 37.8
- 11 Claude Opus 4.8 high 38.1
- 12 Muse Spark Open high 39.1
- 13 GPT-6.1 Sol high 39.9
- 14 Claude Sonnet 4.6 40.4
- 15 Claude Sonnet 4.5 thinking 40.5
- 16 Muse Spark 1.3 high 40.6
- 17 Claude Opus 4.5 thinking 41.6
- 18 GPT-5.5 high 42.4
- 19
Gemini 4 Argon high 42.6 - 20 Claude Opus 4.6 43.2
- 21 GPT-5.6 Sol high 44.5
- 22 GPT-5 high 44.9
- 23 Claude 3.7 Sonnet thinking 46.0
- 24 GPT-5.6 Terra high 46.7
- 25 Claude Sonnet 4 thinking 47.1
- 26 Grok 4.6 48.8
- 27 GLM-5.3 48.8
- 28 GPT-5.4 high 49.2
- 29 Gemini 3.8 Flash high 49.3
- 30 GPT-5.2 high 50.3
- 31 GPT-5.6 Luna high 50.3
- 32 Gemini 3.7 Flash high 51.1
- 33 Grok 4.20 51.7
- 34 GPT-5.1 high 51.8
- 35 GLM-5.2 51.9
- 36 Grok 4.5 53.4
- 37 Kimi K3 high 53.6
- 38 GPT-5.4 Mini high 53.9
- 39 Claude Haiku 4.5 thinking 54.3
- 40 Grok 4.3 54.5
- 41 GPT-5 nano high 55.6
- 42 Gemini 3.1 Pro high 56.5
- 43 GPT-5.4 Nano high 56.8
- 44 GLM-5.1 58.0
- 45 Grok 4 Fast 58.3
- 46 GPT-5 mini high 58.5
- 47 Gemini 2.5 Pro Mar 2025, high 59.0
- 48 Grok 4 59.3
- 49 Kimi K2.6 59.5
- 50 Grok 4.1 Fast 60.0
- 51 Gemini 3.1 Flash-Lite high 61.0
- 52 Gemini 3.5 Flash 61.1
- 53 Kimi K2 Thinking thinking 63.9
- 54 o1 64.0
- 55 Gemini 3 Pro high 64.0
- 56 Gemini 2.5 Pro Jun 2025, high 64.3
- 57 Kimi K2.5 thinking 65.5
- 58
DeepSeek-V4-Pro 65.5 - 59 Gemini 3 Flash high 66.3
- 60 DeepSeek-R1 66.9
- 61 o3-mini high 67.6
- 62 DeepSeek-V3.2 thinking 68.3
- 63 Gemini 2.5 Flash Jun 2025, high 70.0
- 64 GPT-4o Nov 2024 74.4
- 65 Gemini 2.5 Flash-Lite Jun 2025, high 77.1
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 26.2 | high | Sep 2026 | Official leaderboard |
| 2 | o3 | 30.2 | high | Apr 2025 | Official leaderboard |
| 3 | Claude Fable 5.1 | 30.3 | high | Sep 2026 | Official leaderboard |
| 4 | Claude Fable 5 | 30.8 | high | Jun 2026 | Official leaderboard |
| 5 | Muse Wonder | 32.4 | high | Oct 2026 | Official leaderboard |
| 6 | Muse Spark 1.1 | 32.5 | high | Jul 2026 | Official leaderboard |
| 7 | GPT-6 Astra | 35.5 | high | Sep 2026 | Official leaderboard |
| 8 | o4-mini | 35.9 | high | Apr 2025 | Official leaderboard |
| 9 | Claude Opus 5 | 36.1 | high | Jul 2026 | Official leaderboard |
| 10 | Claude Opus 4.7 | 37.8 | Apr 2026 | Official leaderboard | |
| 11 | Claude Opus 4.8 | 38.1 | high | May 2026 | Official leaderboard |
| 12 | Muse Spark Open | 39.1 | high | Oct 2026 | Official leaderboard |
| 13 | GPT-6.1 Sol | 39.9 | high | Sep 2026 | Official leaderboard |
| 14 | Claude Sonnet 4.6 | 40.4 | Feb 2026 | Official leaderboard | |
| 15 | Claude Sonnet 4.5 | 40.5 | thinking | Sep 2025 | Official leaderboard |
| 16 | Muse Spark 1.3 | 40.6 | high | Sep 2026 | Official leaderboard |
| 17 | Claude Opus 4.5 | 41.6 | thinking | Nov 2025 | Official leaderboard |
| 18 | GPT-5.5 | 42.4 | high | Apr 2026 | Official leaderboard |
| 19 | Gemini 4 Argon | 42.6 | high | Sep 2026 | Official leaderboard |
| 20 | Claude Opus 4.6 | 43.2 | Feb 2026 | Official leaderboard | |
| 21 | GPT-5.6 Sol | 44.5 | high | Jul 2026 | Official leaderboard |
| 22 | GPT-5 | 44.9 | high | Aug 2025 | Official leaderboard |
| 23 | Claude 3.7 Sonnet | 46.0 | thinking | Feb 2025 | Official leaderboard |
| 24 | GPT-5.6 Terra | 46.7 | high | Jul 2026 | Official leaderboard |
| 25 | Claude Sonnet 4 | 47.1 | thinking | May 2025 | Official leaderboard |
| 26 | Grok 4.6 | 48.8 | Aug 2026 | Official leaderboard | |
| 27 | GLM-5.3 | 48.8 | Aug 2026 | Official leaderboard | |
| 28 | GPT-5.4 | 49.2 | high | Mar 2026 | Official leaderboard |
| 29 | Gemini 3.8 Flash | 49.3 | high | Sep 2026 | Official leaderboard |
| 30 | GPT-5.2 | 50.3 | high | Dec 2025 | Official leaderboard |
| 31 | GPT-5.6 Luna | 50.3 | high | Jul 2026 | Official leaderboard |
| 32 | Gemini 3.7 Flash | 51.1 | high | Aug 2026 | Official leaderboard |
| 33 | Grok 4.20 | 51.7 | Feb 2026 | Official leaderboard | |
| 34 | GPT-5.1 | 51.8 | high | Nov 2025 | Official leaderboard |
| 35 | GLM-5.2 | 51.9 | Jun 2026 | Official leaderboard | |
| 36 | Grok 4.5 | 53.4 | Jul 2026 | Official leaderboard | |
| 37 | Kimi K3 | 53.6 | high | Jul 2026 | Official leaderboard |
| 38 | GPT-5.4 Mini | 53.9 | high | Mar 2026 | Official leaderboard |
| 39 | Claude Haiku 4.5 | 54.3 | thinking | Oct 2025 | Official leaderboard |
| 40 | Grok 4.3 | 54.5 | Apr 2026 | Official leaderboard | |
| 41 | GPT-5 nano | 55.6 | high | Aug 2025 | Official leaderboard |
| 42 | Gemini 3.1 Pro | 56.5 | high | Feb 2026 | Official leaderboard |
| 43 | GPT-5.4 Nano | 56.8 | high | Mar 2026 | Official leaderboard |
| 44 | GLM-5.1 | 58.0 | Apr 2026 | Official leaderboard | |
| 45 | Grok 4 Fast | 58.3 | Sep 2025 | Official leaderboard | |
| 46 | GPT-5 mini | 58.5 | high | Aug 2025 | Official leaderboard |
| 47 | Gemini 2.5 Pro Mar 2025 | 59.0 | high | Mar 2025 | Official leaderboard |
| 48 | Grok 4 | 59.3 | Jul 2025 | Official leaderboard | |
| 49 | Kimi K2.6 | 59.5 | Apr 2026 | Official leaderboard | |
| 50 | Grok 4.1 Fast | 60.0 | Nov 2025 | Official leaderboard | |
| 51 | Gemini 3.1 Flash-Lite | 61.0 | high | Mar 2026 | Official leaderboard |
| 52 | Gemini 3.5 Flash | 61.1 | May 2026 | Official leaderboard | |
| 53 | Kimi K2 Thinking | 63.9 | thinking | Nov 2025 | Official leaderboard |
| 54 | o1 | 64.0 | Dec 2024 | Official leaderboard | |
| 55 | Gemini 3 Pro | 64.0 | high | Nov 2025 | Official leaderboard |
| 56 | Gemini 2.5 Pro Jun 2025 | 64.3 | high | Jun 2025 | Official leaderboard |
| 57 | Kimi K2.5 | 65.5 | thinking | Jan 2026 | Official leaderboard |
| 58 | DeepSeek-V4-Pro | 65.5 | Apr 2026 | Official leaderboard | |
| 59 | Gemini 3 Flash | 66.3 | high | Dec 2025 | Official leaderboard |
| 60 | DeepSeek-R1 | 66.9 | Jan 2025 | Official leaderboard | |
| 61 | o3-mini | 67.6 | high | Jan 2025 | Official leaderboard |
| 62 | DeepSeek-V3.2 | 68.3 | thinking | Dec 2025 | Official leaderboard |
| 63 | Gemini 2.5 Flash Jun 2025 | 70.0 | high | Jun 2025 | Official leaderboard |
| 64 | GPT-4o Nov 2024 | 74.4 | Nov 2024 | Official leaderboard | |
| 65 | Gemini 2.5 Flash-Lite Jun 2025 | 77.1 | high | Jun 2025 | Official leaderboard |
2 models have held number one since its release in Dec 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
OpenAI Anthropic
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
On the benchmark’s own leaderboard
No lab model card reports it yet
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.
More in Other
- Personal Superintelligence Index (Meta) In use Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO.
- Artificial Analysis Intelligence Index v5 In use Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. Gemini 4 Carbon 59.0
- Anthropic ECI (AECI) Only in system cards Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. Claude Opus 5.5 169.4
- Qwen in-house English automatic evaluation Only in system cards Models answer Qwen in-house English datasets on knowledge, comprehension, coding, math and related skills. Skills and weights change between reports.
Qwen2-72B 72.9% - Recognition of model-written user turns (Anthropic) Only in system cards Models write synthetic user turns for real Claude Code transcripts, then judge which version is real. Score is accuracy above chance. Claude Mythos Preview 0.35
- LiveBench Dropped Models answer monthly refreshed questions across math, coding, reasoning, language, instruction following, and data analysis. Scored automatically against objective ground truth. Claude Fable 5.1 85.7%