| Benchmarking coding agents on Databricks' multi-million-line codebase | No public task-level dataset found | Not stated for benchmark data | Results |
| Warden benchmark | Public aggregate table | Not stated for benchmark results | Results · Dataset or repository |
| LatchBio behavioural-prior study | Public TxBench-PP and scBench-Long JSON summaries and trajectories; aggregate prompt-ablation table in the article | Check the linked benchmark repositories and dataset terms | Results · Dataset 1 · Dataset 2 |
| Harness optimization on the Legal Agent Benchmark | Public article source, harness code, frozen split, dev and test results, and proposer traces | Article text and diagrams are CC BY 4.0; check the linked code and benchmark terms | Results · Dataset or repository |
| OpenBench | Public repository with committed data for several milestones | MIT | Results · Dataset or repository |
| DrugDiscoveryBench | Tasks and code are public; rubrics and reference answers are gated on Hugging Face | MIT for repository code and tasks; separate terms apply to bundled data and gated material | Results · Dataset or repository |
| Artificial Analysis Coding Agent Index | Live public result page and methodology; no stable bulk dataset found | Result-data licence not stated; site terms apply | Results |
| Endor Labs Agent Security League | Public leaderboard and articles | Not stated for benchmark data | Results |
| WANDR benchmark and output-delivery ablation | Public paper, 500 task packages, evaluation code and aggregate tables; evaluated run artefacts are not included | Apache 2.0 repository and CC BY 4.0 paper; third-party task sources retain their own terms | Results · Dataset or repository |
| Vulnerability-Agent-Bench | Public explorer and linked Weights and Biases project | Not stated for benchmark data | Results · Dataset or repository |
| WildClawBench | Public code, tasks, runtime images and dataset | Check the repository and dataset cards | Results · Dataset or repository |
| HarnessBench | Public code and results | Check the repository licence | Results · Dataset or repository |
| CORE-Bench Hard | Public harness, leaderboard and encrypted traces | Check the repository licence | Results · Dataset or repository |
| A-CODE-CLI | Public report and one example task | Check the example repository licence | Results · Dataset or repository |
| The harness tax | Public article with aggregate values | Not stated for request logs | Results |
| GitHub Copilot agentic harness evaluation | Public methods and aggregate charts; no raw result rows found | Not stated for benchmark result data | Results |
| Terminal-Bench 2.1 | Public leaderboard, paper, experiments and submissions | Apache-2.0 repositories; paper CC BY 4.0 | Results · Dataset or repository |
| Agents' Last Exam | Public paper, HTML tables, code and data | Data and paper CC BY 4.0; software Apache-2.0 | Results · Dataset or repository |
| Claw-SWE-Bench | Public paper, adapters, dataset and result artefacts | MIT repository; check dataset card for task data terms | Results · Dataset or repository |
| Harness-Bench: Measuring Harness Effects on LLM Agents | Public paper, tasks, code and aggregated JSON | Paper CC BY 4.0; repository has no licence file as captured | Results · Dataset or repository |
| Harvey LAB harness ablation with Applied Compute | Public aggregate chart; baseline benchmark and harness are public, ablation split is not | MIT for public LAB repository; ablation result-data licence not stated | Results · Dataset or repository |
| Cybersecurity scaffold comparison | Public paper and project page; no dedicated experiment artefact found | Paper CC0; related code has mixed MIT and research-only terms | Results |
| The Scaffold Effect | Public raw results, prompts and analysis | MIT repository | Results · Dataset or repository |
| HAL SWE-bench Verified Mini | Public leaderboard and downloadable traces | Archived trace repository has no licence file as captured | Results · Dataset or repository |
| DeepSeek V4.1 Flash scaffold evaluation | Public model card, technical report, evaluation guide and selected scaffold code; no task-level evaluated results or trajectories | Check the model, harness and benchmark repositories for their separate terms | Results · Dataset 1 · Dataset 2 |
| Prime Agent technical report | Public paper and open-source harness; included long-context values are aggregate-only | MIT for Prime Agent code; check each benchmark for its data terms | Results · Dataset 1 · Dataset 2 · Dataset 3 |
| Browser Use Pi benchmark evidence | Public aggregate reports, configuration manifests, task-level scores, costs and selected audits; tasks and most traces need organization access | MIT repository; private benchmark tasks retain their own access terms | Results · Dataset or repository |