Berkeley Function Calling Leaderboard (BFCL) leaderboard
- Claude Opus 4.5 leads
- 77.5%
- models measured
- 30
- lead over #2
- +4.2 pts
- median
- 47.6%
Leader on Berkeley Function Calling Leaderboard (BFCL) · read October 21
On October 21, 2026, 30 of our catalog's models had a Berkeley Function Calling Leaderboard (BFCL) result, and Anthropic Claude Opus 4.5 led with 77.5%, ahead of Anthropic Claude Sonnet 4.5 at 73.2%.
BFCL tests how accurately a model calls functions, also called tools. Its sub-categories include single-turn calls, live functions contributed by users, multi-turn conversations, web search and memory. The number is overall accuracy, the unweighted average of all the sub-categories, as Berkeley publishes it. Models are listed in function-calling mode or prompt mode; we show the better of the two.
Role in the index
Berkeley Function Calling Leaderboard (BFCL) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5Anthropic | 77.5% | 44.0 | $25.00 | |
| 2 | Claude Sonnet 4.5Anthropic | 73.2% | 39.5 | $15.00 | |
| 3 | GLM 4.6Z.ai | 72.4% | 8.8 | $2.20 | |
| 4 | Claude Haiku 4.5Anthropic | 68.7% | 35.6 | $5.00 | |
| 5 | o3OpenAI | 63.0% | Not benchmarked | $8.00 | |
| 6 | DeepSeek V3.2 ExpDeepSeek | 56.7% | Not benchmarked | $0.41 | |
| 7 | Gemini 2.5 FlashGoogle | 56.2% | Not benchmarked | $2.50 | |
| 8 | GPT-5.2OpenAI | 55.9% | 55.9 | $14.00 | |
| 9 | GPT-5 MiniOpenAI | 55.5% | 34.2 | $2.00 | |
| 10 | GPT-4.1OpenAI | 54.0% | Not benchmarked | $8.00 | |
| 11 | o4 MiniOpenAI | 53.2% | Not benchmarked | $4.40 | |
| 12 | Qwen3 235B A22B Instruct 2507Qwen | 52.1% | Not benchmarked | $0.75 | |
| 13 | GPT-5 NanoOpenAI | 51.4% | Not benchmarked | $0.40 | |
| 14 | GPT-4.1 MiniOpenAI | 50.4% | Not benchmarked | $1.60 | |
| 15 | Qwen3 32BQwen | 48.7% | Not benchmarked | $0.40 | |
| 16 | Command ACohere | 46.5% | Not benchmarked | $10.00 | |
| 17 | Qwen3 30B A3B Instruct 2507Qwen | 41.4% | Not benchmarked | $0.30 | |
| 18 | Qwen3 14BQwen | 41.0% | Not benchmarked | $0.23 | |
| 19 | Gemini 2.5 Flash LiteGoogle | 36.9% | Not benchmarked | $0.40 | |
| 20 | GPT-4.1 NanoOpenAI | 33.1% | Not benchmarked | $0.40 | |
| 21 | Command R7B (12-2024)Cohere | 32.1% | Not benchmarked | $0.15 | |
| 22 | Llama 3.3 70B InstructMeta Llama | 31.9% | Not benchmarked | $0.72 | |
| 23 | Gemma 3 12BGoogle | 30.4% | Not benchmarked | $0.1625 | |
| 24 | Gemma 3 27BGoogle | 29.5% | Not benchmarked | $0.30 | |
| 25 | Phi 4Microsoft | 28.8% | Not benchmarked | $0.14 | |
| 26 | Llama 4 ScoutMeta Llama | 28.1% | Not benchmarked | $0.50 | |
| 27 | Llama 3.1 8B InstructMeta Llama | 25.8% | Not benchmarked | $0.08 | |
| 28 | Llama 3.2 3B InstructMeta Llama | 21.9% | Not benchmarked | $0.3325 | |
| 29 | Gemma 3 4BGoogle | 19.6% | Not benchmarked | $0.075 | |
| 30 | Llama 3.2 1B InstructMeta Llama | 10.8% | Not benchmarked | $0.201 |
Only models in our catalog are listed, and each appears once with its best published run. 30 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out.