Results
Results
Every published run, newest first, each with its own pass rate. Filter it, or download what you see as CSV.
Select two rows of one board to compare them task by task.
| Harness | Board | Pass rateHigher is better, 95% interval | Run | Cost per task | hbench | Finished |
|---|---|---|---|---|---|---|
| Terminal-Bench 2.0 on gpt-5.2 | 60.7%50.6–70.8 | #73 | — | converted | 16 Jan 2026 | |
| Terminal-Bench 2.0 on claude-opus-4-5 | 58.4%48.3–68.5 | #71 | — | converted | 16 Jan 2026 | |
| Terminal-Bench 2.0 on claude-opus-4-5 | 51.7%41.6–61.8 | #77 | — | converted | 12 Jan 2026 | |
| Terminal-Bench 2.0 on Gemini 3 Pro Preview (5 attempts, open network) | 64.7%56.2–72.1 | #43 | — | converted | 1 Jan 2026 | |
aone-agent1.18.72-pre | Terminal-Bench 2.0 on qwen3-coder-480b-a35b-instruct-fp8 (5 attempts, open network) | 27.2%19.8–35.1 | #93 | — | converted | 26 Dec 2025 |
final1.5-official | Terminal-Bench 2.0 on Claude Sonnet 4.5 (5 attempts, open network) | 42.7%32.6–52.8 | #59 | — | converted | 9 Dec 2025 |