Benchmark · Agent tasks
Berkeley Function Calling Leaderboard (BFCL)
Whether a model calls functions correctly: simple, parallel and multi-turn calls, knowing when no function fits, web search and memory tasks, run by the Berkeley team on every model, with native function calling and with prompting as separate rows.
- Score
- % overall accuracy, higher is better
- Run by
- UC Berkeley (Gorilla)
- Status
- Stale, newest result 16 December 2025, nothing new in six months
- Updated
- irregular
- Data
- the published file, Apache-2.0, read 20 November 2026
Results
5 models we list, each at its best setting, best first.
| # | Model | Score | Range | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3-30B-A3B-Instruct-2507 C | 41.4% | Independent | mode FC | 2025-12-16 | source | |
| 2 | Qwen3-4B-Instruct-2507 C | 35.7% | Independent | mode FC | 2025-12-16 | source | |
| 3 | Command R7B B | 32.1% | Independent | mode FC | 2025-12-16 | source | |
| 4 | Llama 3.3 70B B | 31.9% | Independent | mode FC | 2025-12-16 | source | |
| 5 | Amazon Nova 2 Lite A | 27.1% | Independent | mode FC | 2025-12-16 | source |
Every setting, 7 rows
| Model | As the leaderboard names it | Score | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|
| Qwen3-30B-A3B-Instruct-2507 | Qwen3-30B-A3B-Instruct-2507 | 41.4% | Independent | mode FC | 2025-12-16 | source |
| Qwen3-30B-A3B-Instruct-2507 | Qwen3-30B-A3B-Instruct-2507 | 36.7% | Independent | mode Prompt | 2025-12-16 | source |
| Qwen3-4B-Instruct-2507 | Qwen3-4B-Instruct-2507 | 35.7% | Independent | mode FC | 2025-12-16 | source |
| Qwen3-4B-Instruct-2507 | Qwen3-4B-Instruct-2507 | 35.5% | Independent | mode Prompt | 2025-12-16 | source |
| Command R7B | Command R7B | 32.1% | Independent | mode FC | 2025-12-16 | source |
| Llama 3.3 70B | Llama-3.3-70B-Instruct | 31.9% | Independent | mode FC | 2025-12-16 | source |
| Amazon Nova 2 Lite | Amazon-Nova-2-Lite-v1:0 | 27.1% | Independent | mode FC | 2025-12-16 | source |
A rank is among the models we list with a result, not a place on the whole leaderboard, and a model offered by several hosts is shown once, on its own listing. These results don't move our grades. Data: Berkeley Function Calling Leaderboard, UC Berkeley (Gorilla), Apache 2.0, Apache-2.0.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.