Kurakurai French LLM - BFCL v2: leaderboard
Evaluates function and tool calling in French on the Kurakurai French LLM leaderboard, a static board of small models up to about 4B parameters. The score is BFCL v2 Overall: the unweighted mean of the Non-Live and Live halves, where Non-Live averages five categories unweighted and Live is count-weighted over six. It is computed on a cleaned French translation of 3,638 single-turn samples across 13 categories, with no French multi-turn layer, and averaged over three sampling seeds rather than the ten used by the board's other columns. Upstream hides this column by default and excludes it from the board's visible-benchmark mean, since few models have a score. This series reports Kurakurai French LLM - BFCL v2 using Accuracy (%).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 4B | 73.61 | #514 |
| 2 | Gemma 4 E2B (IT) | 72.98 | #779 |
| 3 | Granite 4.1 3B | 72.75 | #899 |
| 4 | granite-4.0-h-micro | 72.3 | #1078 |
| 5 | SmolLM3-3B | 63.55 | #1239 |
| 6 | Luth-0.6B-Instruct | 61.72 | #1402 |
| 7 | LFM2 2.6B | 61.27 | #1313 |
| 8 | LFM2.5-1.2B Instruct | 60.38 | #1451 |
| 9 | Qwen 3 4B 2507 Instruct | 57.44 | #768 |
| 10 | Llama 3.2 3B Instruct | 55.53 | #1327 |
| 11 | Granite 4.0 H 350M | 55.38 | #1502 |
| 12 | Qwen 3 1.7B | 53.4 | #1175 |
| 13 | Qwen 3.5 2B | 51.78 | #1114 |
| 14 | Qwen 3.5 0.8B | 51.49 | #1410 |
| 15 | LFM2-700M | 51.17 | #1537 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kurakurai-french-llm-bfcl-v2 · How It Works · Data refreshed daily, snapshot 2026-11-05.