AgentHorizon
Harbor: pendingA benchmark of 1,373 computer-use tasks with paired instructions to evaluate agentic judges on long-horizon trajectory success verification across three OSs.
- Samples
- 1,373
- Released
- 2026
- Scores
- 10
ExploitBench
ExploitBench: Capability Ladder Benchmark for LLM Cybersecurity Agents (V8 exploit development)
ExploitBench evaluates LLM agents on developing memory-corruption exploits against 41 real V8 JavaScript engine N-day vulnerabilities with all production mitigations enabled.
How to run
ltb · exploitbench
$ npm i -g lettokenburn-cli$ ltb bench pull exploitbench
External harness. Externally registered Inspect eval. Install from https://github.com/ChaoticCooties/exploitbench-eval and run `inspect eval eb_v8 --model <provider/model>` (see the repo README). https://github.com/ChaoticCooties/exploitbench-eval
License is link-only: ltb bench pull fetches rows from the original host for local evaluation only and does not write a solution/ directory. Do not redistribute the generated tasks.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
10 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Mythos Preview Early | 73.8 | N/A | Independent | low | N/A | epoch.ai |
| GPT-5.5 | 47.4 | N/A | Independent | low | N/A | epoch.ai |
| GPT-5.5 | 41.6 | Codex | Independent | low | N/A | epoch.ai |
| Claude Opus 4.7 | 26.5 | N/A | Independent | low | N/A | epoch.ai |
| Gemini 3.1 Pro | 26.1 | N/A | Independent | low | N/A | epoch.ai |
| Claude Sonnet 4.6 | 23.6 | N/A | Independent | low | N/A | epoch.ai |
| Kimi K2.6 | 18.4 | N/A | Independent | low | N/A | epoch.ai |
| GLM-5.1 | 18.1 | N/A | Independent | low | N/A | epoch.ai |
| Claude Haiku 4.5 | 13.7 | N/A | Independent | low | N/A | epoch.ai |
| MiniMax-M2.7 | 13.3 | N/A | Independent | low | N/A | epoch.ai |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
ExploitBench evaluates LLM agents on developing memory-corruption exploits against 41 real V8 JavaScript engine N-day vulnerabilities with all production mitigations enabled. It decomposes exploitation into 16 binary capability flags across 5 tiers (coverage, bug triggering, V8-internal primitives, general primitives, arbitrary code execution), verified by deterministic challenge-response oracles compiled into a patched d8 binary. Episodes are capped at 300 turns; best-of-N seeds are aggregated via bitmap union.