Compare how leading models behave across the risks that matter most.
A composite measure of how widely used AI models perform across safety risks consumers may encounter.
Showing 14 of 14 model configurations.
| Model | Provider | Model version | Reasoning setting | Score out of 100 | Interval | Source | Measured | Sample |
|---|---|---|---|---|---|---|---|---|
| GPT-6 Astra · Medium | OpenAI | GPT-6 Astra | Medium | 70.15 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Grok 4.6 · Medium | xAI | Grok 4.6 | Medium | 68.80 | Not supplied | Weighted mean | Not supplied | Not supplied |
| GPT-6 Sol · Medium | OpenAI | GPT-6 Sol | Medium | 67.86 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Perplexity Agent · medium preset | Perplexity | Perplexity Agent · medium preset | Not recorded | 67.58 | Not supplied | Weighted mean | Not supplied | Not supplied |
| GPT-5.6 Terra · Medium | OpenAI | GPT-5.6 Terra | Medium | 67.53 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Claude Opus 5.5 · Medium | Anthropic | Claude Opus 5.5 | Medium | 66.06 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Grok 4.5 · Medium | xAI | Grok 4.5 | Medium | 65.41 | Not supplied | Weighted mean | Not supplied | Not supplied |
| GPT-5.6 Luna · Medium | OpenAI | GPT-5.6 Luna | Medium | 65.37 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Claude Sonnet 5 · Medium | Anthropic | Claude Sonnet 5 | Medium | 64.84 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Inkling · Medium | Inkling | Inkling | Medium | 64.30 | Not supplied | Weighted mean | Not supplied | Not supplied |
| GLM 5.3 Flash · High | Zhipu AI | GLM 5.3 Flash | High | 63.23 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Gemini 3.6 Flash · Medium | Gemini 3.6 Flash | Medium | 60.38 | Not supplied | Weighted mean | Not supplied | Not supplied | |
| DeepSeek V4 Flash · Medium | DeepSeek | DeepSeek V4 Flash | Medium | 55.65 | Not supplied | Weighted mean | Not supplied | Not supplied |
| Mistral Medium 3.5 | Mistral AI | Mistral Medium 3.5 | Not recorded | 45.42 | Not supplied | Weighted mean | Not supplied | Not supplied |
We believe AI can and will improve human life. But what happens when two billion people interact with imperfect AI systems every single day?
From health, relationship, financial, and legal advice to suicidal ideation and racial bias, AI’s growing role in personal decision making demands independent, systematic evaluation of risks and safety standards.
The AI Safety Index brings those findings together, translating technical evidence into clear, accessible data.
See where each model performs well and where it falls short.
How well models respond to emotional distress without reinforcing delusions or encouraging dangerous behavior.
How safely models respond to children & teens asking about sensitive or risky topics.
How well models answer health questions while avoiding advice that could cause harm.
How well models help you make informed decisions without flattery, pressure, or hidden persuasion.
How well models recognize scam warning signs and alert you.
How well models avoid stereotypes and biased answers.
How well models avoid sharing your private information with the wrong people.
How well models get facts right, correct false claims, and resist pressure to agree.
How reliably models follow instructions and respect boundaries they have been given.
The latest additions to the index.
· New model results
· New model results
· New model results
· New model results
· New model results
The index stands on benchmarks built by researchers at these institutions.
Hover over a logo to see which benchmarks its researchers built.Tap a logo to see which benchmarks its researchers built.
Join the mailing list to hear what’s next.