Horangi 4 - BFCL: leaderboard
Berkeley Function Calling Leaderboard categories (simple, multiple, Java, JavaScript, live and relevance) as sampled by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Accuracy (%), higher is better. Runs whose logged breakdown is degenerate (every call category 0, relevance 1) are omitted.
Metric: Accuracy (%). Source: horangi.ai. Saturation forecast: Not forecast. 81 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5 (High) | 71.71 | #65 (Claude Sonnet 5) |
| 2 | Gemma 4 E4B (IT) (Thinking) | 70.95 | |
| 3 | A.X 4.0 | 70.54 | #436 |
| 4 | GLM-5V Turbo | 70.16 | #172 |
| 5 | GLM-5-Turbo | 70.16 | #121 |
| 6 | Claude Opus 4.5 (20251101) (High) | 70.16 | #81 (Claude Opus 4.5 (20251101)) |
| 7 | Claude Sonnet 4.5 (High) | 69.77 | #148 (Claude Sonnet 4.5) |
| 8 | Gemma 4 E2B (IT) (Thinking) | 69.32 | #779 (Gemma 4 E2B (IT)) |
| 9 | Qwen 3.5 122B A10B | 68.99 | #184 |
| 10 | GLM-5 | 68.99 | #149 |
| 11 | Claude Sonnet 4.6 (High) | 68.99 | #94 (Claude Sonnet 4.6) |
| 12 | gemma-4-26B-A4B-it (Thinking) | 68.7 | |
| 13 | Kimi K2.5 | 68.6 | #152 |
| 14 | Grok 4.20 | 68.6 | #179 |
| 15 | Gemini 3.1 Pro (Preview) (High) | 68.6 | #64 (Gemini 3.1 Pro (Preview)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-bfcl · How It Works · Data refreshed daily, snapshot 2026-11-05.