Anthropic Binary Exploitation Benchmark (formerly OSS-Fuzz), control-flow hijacks (Binary Exploitation Benchmark)
We show this table for reference; we do not rank on it.
Anthropic's unguided vulnerability-discovery and exploitation evaluation: from a single function entry point in an open-source project, the model must find a bug and push it toward a working exploit. This lane counts control-flow hijacks.
Benchmark score on Binary Exploitation Benchmark — November 11, 2026
We compile the Binary Exploitation Benchmark rows from provider self-reports. Claude Mythos 5.2 leads the table at 148, followed by Claude Opus 5.5 (112) and Claude Mythos 5.1 (84). We do not use these results to rank models overall.
Claude Mythos 5.2
Anthropic
Claude Opus 5.5
Anthropic
Claude Mythos 5.1
Anthropic
4 modelsAgenticCurrentDisplay onlyUpdated November 11, 2026
Benchmark score table (4 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Mythos 5.2Anthropic | 148 | Not reported | Closed |
| 2 | Claude Opus 5.5Anthropic | 112 | Not reported | Closed |
| 3 | Claude Mythos 5.1Anthropic | 84 | Not reported | Closed |
| 4 | Claude Sonnet 5.5Anthropic | 48 | Not reported | Closed |
Among the reported Binary Exploitation Benchmark rows, Claude Mythos 5.2 is first at 148. The third row is 64 score units behind. The broader top-10 range is 100 score units, so the table still separates the published systems.
4 models have been evaluated on Binary Exploitation Benchmark. The benchmark falls in the Agentic category. Binary Exploitation Benchmark is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Binary Exploitation Benchmark
Year
2026
Tasks
Open-source function entry points
Format
Count of control-flow hijacks
Difficulty
Exploit development
Run with cyber mitigations off. Earlier system cards reported this evaluation as OSS-Fuzz crash and write-primitive rates, stored on separate lanes; the Fable 5.2 card reports counts of control-flow hijacks. Dual-use offensive capability, so display-only.
Freshness and provenance
Version
Binary Exploitation Benchmark 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Binary Exploitation Benchmark measure?
Anthropic's unguided vulnerability-discovery and exploitation evaluation: from a single function entry point in an open-source project, the model must find a bug and push it toward a working exploit. This lane counts control-flow hijacks.
Which model scores highest on Binary Exploitation Benchmark?
Claude Mythos 5.2 by Anthropic currently leads with a score of 148 on Binary Exploitation Benchmark.
How many models are evaluated on Binary Exploitation Benchmark?
4 AI models have published results on Binary Exploitation Benchmark in the BenchLM catalog.
Compare top models on Binary Exploitation Benchmark
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.