Category ranking · 7 models · checked 2026-10-13
Tool use and agents
Ranked by BFCL V4, overall accuracy. Source: Berkeley Function-Calling Leaderboard data_overall.csv.
BFCL V4, overall accuracy
3 with a published score, 4 not yet measured.
# | Model | Parameters | Score | Source | Licence |
|---|---|---|---|---|---|
| 1 | xLAM-2 8B Salesforce | 8.0 B | 46.68%Sourced | Source xLAM-2-8b-fc-r (FC) row, rank 34. | CC-BY-NC-4.0 Check terms before commercial use |
| 2 | ToolACE-2 8B Team ACE | 8.0 B | 42.44%Sourced | Source ToolACE-2-8B (FC) row, rank 40. | Apache-2.0 |
| 3 | Qwen3 4B Instruct 2507 Alibaba Qwen | 4.0 B | 35.68%Sourced | Source Qwen3-4B-Instruct-2507 (FC) row, rank 54. | Apache-2.0 |
| — | Arch-Function 3B Katanemo | 3.1 B | Not yet measured | — | Katanemo Research Check terms before commercial use |
| — | GLM-4.5 Air Z.ai | 110.5 B | Not yet measured | Not listed in BFCL V4. | MIT |
| — | Hermes 3 8B Nous Research | 8.0 B | Not yet measured | — | Llama 3 Community |
| — | watt-tool 8B Watt AI | 8.0 B | Not yet measured | — | Apache-2.0 |
How to read this: one public benchmark per category, so you can compare like with like. We copy the published number and link it; we do not re-run it, average different benchmarks or fill gaps. Vendor-reported numbers use each vendor's own settings, so treat small gaps as ties. “Not yet measured” means we found no comparable public score, not that the model is weak.
Every model was checked against the Hugging Face API on 2026-10-13: it exists, is public, and has the parameter count and licence shown. Other categories: Math and reasoning · Coding · Vision-language · OCR and documents · Speech to text · Text to speech · Image generation · Embeddings and rerankers · Small fast 'System 1' · Biomedical · Multilingual and translation · Time series · Robotics (VLA).
Machine-readable: /api/v1/categories/tool-use-agents.json