Multimodal Agentic Tool Use on TerminalBench-O
24Pass RateGPT-5.5 Low
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5.5 LowHarness=Codex2026.05 | 24 | 0.68 | |
| GPT-5.5 HighHarness=Codex2026.05 | 24 | 0.71 | |
| GPT-5.5 xHighHarness=Codex2026.05 | 24 | 0.71 | |
| GPT-5.5 MediumHarness=Codex2026.05 | 18 | 0.69 | |
| GLM-5.1 (V)Harness=CC, Native Vision=No (uses image-read with Qwen3.5-Flash)2026.05 | 16 | 0.69 | |
| DeepSeek V4 Pro (V)Harness=CC, Native Vision=No (uses image-read with Qwen3.5-Flash)2026.05 | 12 | 0.64 | |
| DeepSeek V4 Flash (V)Harness=CC, Native Vision=No (uses image-read with Qwen3.5-Flash)2026.05 | 10 | 0.6 | |
| Kimi K2.6Harness=CC2026.05 | 8 | 0.56 | |
| MiniMax M2.7 (V)Harness=CC, Native Vision=No (uses image-read with Qwen3.5-Flash)2026.05 | 4 | 0.55 |