Berkeley Function-Calling Leaderboard: leaderboard
Standardized evaluation of LLM function-calling abilities with AST correctness and accuracy metrics across API schemas.
Metric: Overall Accuracy (%). Source: gorilla.cs.berkeley.edu. Status: saturation imminent. 109 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 77.47 |
| 2 | Claude Sonnet 4.5 | 73.24 |
| 3 | Gemini 3 Pro (Preview) | 72.51 |
| 4 | GLM-4.6 (Thinking) | 72.38 |
| 5 | Grok 4.1 Fast (Reasoning) | 69.57 |
| 6 | Claude Haiku 4.5 (20251001) | 68.7 |
| 7 | O3 (2025-04-16) | 63.05 |
| 8 | Grok 4 | 62.97 |
| 9 | Kimi K2 | 59.06 |
| 10 | Grok 4.1 Fast (Non-reasoning) | 58.29 |
| 11 | DeepSeek V3.2 Exp (Thinking) | 56.73 |
| 12 | Gemini 2.5 Flash | 56.24 |
| 13 | GPT-5.2 | 55.87 |
| 14 | GPT-5 Mini | 55.46 |
| 15 | DeepSeek V3.2 Exp | 54.12 |
Interactive version: theaggregate.ai/benchmark?slug=berkeley-function-calling-leaderboard · How It Works · Data refreshed daily, snapshot 2026-09-05.