Firefox 147 exploit evaluation (Anthropic)
Internal eval on crashes previously found in Firefox 147, run without the browser's process sandbox. Measures exploit-development progress.
- 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
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Claude Mythos Preview leads with 72.4% among models released by Jul 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Jul 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 14 more
- 1 Claude Mythos Preview 72.4%
- 2 Claude Opus 5 52.4%
- 3 Claude Opus 4.7 45.2%
- 4 Claude Sonnet 5 13.2%
- 5 Claude Opus 4.8 8.80%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Mythos Preview | 72.4% | Apr 2026 | Model card | |
| 2 | Claude Opus 5 | 52.4% | Jul 2026 | Model card | |
| 3 | Claude Opus 4.7 | 45.2% | Apr 2026 | Model card | |
| 4 | Claude Sonnet 5 | 13.2% | Jun 2026 | Model card | |
| 5 | Claude Opus 4.8 | 8.80% | May 2026 | Model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
5 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 24 Jul 2026 | Claude Opus 5 system card | 52.4 | Firefox 147, full working exploits in 131/250 trials, partial progress in 218 (87.2%), mitigations off (3.3.3) | Cyber risk |
| 30 Jun 2026 | Claude Sonnet 5 system card | 13.2 | Firefox 147 exploit development, any-success (register control or better) 33/250 trials, full working exploit 0.0%, mitigations off (0 with mitigations) (3.2.4, Figure 3.2.4.A) | Cyber risk |
| 28 May 2026 | Claude Opus 4.8 system card | 8.8 | Firefox 147 JS shell exploitation, full working exploit (grade 1.0) in 22 of 250 trials, at least register control 68.8%, safeguards off. No points with safeguards on (3.3.3, Figure 3.3.3.A) | Cyber risk |
| 16 Apr 2026 | Claude Opus 4.7 system card | 45.2 | Firefox 147 JS shell exploitation, success rate with partial control or better over 250 trials, printed on chart. Full code execution 1.2% (3.3.3, Figure 3.3.3.A) | Cyber risk |
| 7 Apr 2026 | Claude Mythos Preview system card | 72.4 | Firefox 147 JS shell exploitation, full code execution rate over 250 trials, value printed on chart; 84.0 including partial control. Variant without the top 2 bugs: 4.4 full, 85.2 including partial (3.3.3, Figures 3.3.3.A and 3.3.3.B) | Cyber risk |
More in Cybersecurity
- ExploitGym In use 869 real-vulnerability exploit-development tasks (userspace, V8, Linux kernel). Success means capturing a flag through the intended vulnerability. GPT-6 Astra 42.4%
- ExploitBench In use 41 recent V8 JavaScript-engine N-day vulnerabilities. Progress measured by 16 grader-verified capability flags across five tiers. GPT-6 Astra 100%
- CWE-bench In use Agents audit a repository and write a patch for a security flaw, without being told where the bug is. Covers six programming languages.
Gemini 4 Carbon 71% - Real-world vulnerability discovery In use Agents search popular open-source codebases to rediscover about 1,200 recent confirmed vulnerabilities across many languages. Reports share discovered. Gemini 4 Argon 86%
- Artificial Analysis Cyber Index In use Agents do enterprise cyber defense tasks from three benchmarks, CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA, combined into one success rate.
Mistral Large 4 (Le Chonk, open weights) 51% - Big Sleep Evaluation (Google) In use Agents search large real codebases such as Chrome and Safari for critical, hard to find vulnerabilities. Scored by pass@1 success rate, run without safety guardrails. Gemini 3.5 Flash Cyber 72%