MolBench-MO Molecule Editing (39 tasks)
100AccuracyMolClaw (Claude Code)
Evaluation Results
| Method | Links | |
|---|---|---|
| MolClaw (Claude Code)Category=Agent, LLM backend=Claude Sonnet 4.6, Platform=Claude Code2026.04 | 100 | |
| Claude Code (vanilla)Category=Agent, LLM backend=Claude Sonnet 4.62026.04 | 97.4 | |
| OpenClaw (vanilla)Category=Agent, LLM backend=Claude Sonnet 4.62026.04 | 97.4 | |
| MolClaw (OpenClaw)Category=Agent, LLM backend=Claude Sonnet 4.6, Platform=OpenClaw2026.04 | 97.4 | |
| Claude Sonnet 4.6Category=LLM2026.04 | 94.9 | |
| Qwen 3.5Category=LLM2026.04 | 90 | |
| Kimi 2.5Category=LLM2026.04 | 89.7 | |
| GPT 5.2Category=LLM2026.04 | 84.6 | |
| GLM 5Category=LLM2026.04 | 74.4 | |
| Gemini 3Category=LLM2026.04 | 51 | |
| Minimax 2.5Category=LLM2026.04 | 30.8 | |
| Deepseek V3.2Category=LLM2026.04 | 22.5 | |
| BiomniCategory=Agent2026.04 | 0 |