BlogLLM Benchmark Scores as Published, Side by Side
ExploitBench: Results as Published
Every result the catalog holds for ExploitBench: 26 models from 7 developers, read from 11 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area CybersecurityMaintainer Seunghyun Lee and David BrumleyIntroduced 2026Data version 2026103001, as of 2026-10-30
ExploitBench measures how far an AI agent gets along a ladder of 16 checkable capability steps when it works on known, already-patched bugs in the V8 JavaScript engine used by Chromium. The steps run from triggering the bug up to full control of the target, and each step is confirmed by an automatic verifier with no AI judge. The public set has 41 bugs, all reported in 2024 or later.
ExploitBench
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6.1 TerraOpenAI, released 2026-10-26 | 97.1% | OpenAI, GPT-6.1 Terra system card addendumchecked 2026-10-26 | - |
| GPT-6.1 SolOpenAI, released 2026-09-29 | 99.7% | OpenAI, GPT-6.1 Sol system card addendumchecked 2026-10-19 | - |
| GPT-6 LunaOpenAI, released 2026-09-22 | 44.66% | OpenAI, GPT-6 Sol and GPT-6 Luna: October 2026 updatechecked 2026-10-19 | - |
| GPT-6 SolOpenAI, released 2026-09-22 | 82.62% | OpenAI, GPT-6 Sol and GPT-6 Luna: October 2026 updatechecked 2026-10-19; printed on 1 more page | - |
| GPT-6 AstraOpenAI, released 2026-09-03 | 100% | OpenAI, GPT-6 Astrachecked 2026-10-14; printed on 2 more pages | - |
| GPT-5.6 LunaOpenAI, released 2026-07-09 | 33.2% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Harness: ExploitBench API harness, 5 seeds, reasoning continuity (footnote 2) |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 73.5% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15; printed on 3 more pages | Harness: ExploitBench API harness, 5 seeds, reasoning continuity (footnote 2) |
| GPT-5.6 TerraOpenAI, released 2026-07-09 | 52.9% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Harness: ExploitBench API harness, 5 seeds, reasoning continuity (footnote 2) |
| GPT-5.5OpenAI, released 2026-04-23 | 47.9% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15; printed on 1 more page | Harness: ExploitBench API harness, 5 seeds, reasoning continuity (footnote 2) |
| Claude Haiku 5.5Anthropic, released 2026-10-07 | 38% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: ExploitBench, static uniform harness, all cyber safeguards off, escape classifiers on (Figure 3.3.1.A); plain arm |
| 49% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: ExploitBench, static uniform harness, all cyber safeguards off, escape classifiers on (Figure 3.3.1.A); AutoNudge arm | |
| Claude Opus 5.5Anthropic, released 2026-09-22 | 91% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Fallback: with fallback |
| Claude Fable 5.1Anthropic, released 2026-09-01 | 83% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Fallback: with fallback |
| Claude Opus 5Anthropic, released 2026-07-24 | 70% | OpenAI, GPT-6 Astrachecked 2026-10-14; printed on 1 more page | - |
| Claude Fable 5Anthropic, released 2026-06-09 | 78% | Xiaomi, Xiaomi MiMo-V2.6-Pro-RL model cardchecked 2026-10-19; printed on 1 more page | - |
| Claude Mythos 5Anthropic, released 2026-06-09 | 78% | Anthropic, Claude Fable 5 and Claude Mythos 5checked 2026-10-15; printed on 1 more page | - |
| Claude Opus 4.8Anthropic, released 2026-05-28 | 40% | Anthropic, Claude Fable 5 and Claude Mythos 5checked 2026-10-15; printed on 2 more pages | - |
| Claude Mythos PreviewAnthropic, released 2026-04-07 | 69% | Anthropic, Claude Fable 5 and Claude Mythos 5checked 2026-10-15; printed on 1 more page | - |
| Muse Wonder FullMeta, released 2026-10-12 | 58.9% | Meta, Meta Muse Wonder model cardchecked 2026-10-19 | - |
| Qwen3.8-MaxAlibaba, released 2026-08-03 | 28.8% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| Kimi K3.1 ProMoonshot AI, released 2026-10-23 | 74.7% | Moonshot AI, Introducing Kimi K3.1 Prochecked 2026-10-24 | Effort: max; Harness: ExploitBench official harness; Attempt budget: 300 turns; pass@5 |
| Kimi K3Moonshot AI, released 2026-07-16 | 32.2% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| GLM-5.3Z.ai, released 2026-08-18 | 54.4% | Z.ai, GLM-5.3 model cardchecked 2026-10-15; printed on 1 more page | Effort: max; Harness: Claude Code 2.1.207 |
| GLM-5.2Z.ai, released 2026-06-16 | 24.4% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| MiMo-V2.6-FlashXiaomi, released 2026-09-21 | 25.3% | Xiaomi, Xiaomi MiMo-V2.6-Pro-RL model cardchecked 2026-10-19 | - |
| MiMo-V2.6-ProXiaomi, released 2026-09-21 | 47.9% | Xiaomi, Xiaomi MiMo-V2.6-Pro-RL model cardchecked 2026-10-19 | - |
| MiMo-V2.5-ProXiaomi, released 2026-04-27 | 16.6% | Xiaomi, Xiaomi MiMo-V2.6-Pro-RL model cardchecked 2026-10-19 | - |
ExploitBench (June-Aug 2026)
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6.1 TerraOpenAI, released 2026-10-26 | 24.5% | OpenAI, GPT-6.1 Terra system card addendumchecked 2026-10-26 | - |
| GPT-6 AstraOpenAI, released 2026-09-03 | 39% | OpenAI, GPT-6 Astrachecked 2026-10-14 | - |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 5.5% | OpenAI, GPT-6 Astrachecked 2026-10-14 | - |
ExploitBench full arbitrary code execution
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Claude Haiku 5.5Anthropic, released 2026-10-07 | 1% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
ExploitBench mean capability flags
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Claude Haiku 5.5Anthropic, released 2026-10-07 | 5.15 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: ExploitBench, static uniform harness, all cyber safeguards off, escape classifiers on (Figure 3.3.1.A); plain arm |
| 6.56 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: ExploitBench, static uniform harness, all cyber safeguards off, escape classifiers on (Figure 3.3.1.A); AutoNudge arm |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.