Benchmark Almanac

BFCL V4

Qwen3.7-Max 75% leads since May 2026 In use

Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.

V4 added agentic web search and memory categories and changed the overall formula, so its scores do not compare with V3.

Made by UC Berkeley (Gorilla) Released Jul 2025 Function calling All tasks public Last in a launch post Oct 2026 Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
ShishirPatil/gorilla 13,062 stars Leaderboard
24
Models scored in lab cards
25
Launch posts list it, from 12 labs
11
System cards report it, from 7 labs

Qwen3.7-Max leads with 75%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

24 models. Scroll sideways for the rest.

  1. 75.0 Qwen3.7-Max
  2. 73.4 Seed2.0 Pro
  3. 72.9 Qwen3.5 397B-A17B
  4. 72.9 Qwen3.7-Plus
  5. 72.2 Qwen3.5-122B-A10B
  6. 71.8 Spark-X2.6-4B
  7. 71.4 LFM3-1.7B
  8. 70.6 Ling-2.6-1T
  9. 70.1 Trinity Large Thinking
  10. 67.7 Qwen3-Max
  11. 66.5 ERNIE 5.0
  12. 65.7 Keye-VL-2.0-30B-A3B
  13. 65.0 LFM2.5-24B-A2B
  14. 63.3 Qwen3.5-Omni-Plus
  15. 61.6 Nova 2.0 Pro Preview
  16. 61.4 Granite 4.2 30B
  17. 60.3 Nova 2 Lite
  18. 58.3 Nova 2 Omni Preview
  19. 57.2 Seed1.8
  20. 56.9 LFM2.5-2.6B
  21. 53.8 Nemotron 3 Nano 30B-A3B
  22. 52.9 Nemotron-Cascade 2 30B-A3B
  23. 48.5 LFM2.5-8B-A1B
  24. 32.5 LFM2.5-VL-3B
  1. 1 Qwen3.7-Max 75%
  2. 2 Seed2.0 Pro 73.4%
  3. 3 Qwen3.5 397B-A17B 72.9%
  4. 4 Qwen3.7-Plus 72.9%
  5. 5 Qwen3.5-122B-A10B 72.2%
  6. 6 Spark-X2.6-4B 71.8%
  7. 7 LFM3-1.7B 71.4%
  8. 8 Ling-2.6-1T 70.6%
  9. 9 Trinity Large Thinking 70.1%
  10. 10 Qwen3-Max 67.7%
  11. 11 ERNIE 5.0 66.5%
  12. 12 Keye-VL-2.0-30B-A3B 65.7%
  13. 13 LFM2.5-24B-A2B 65%
  14. 14 Qwen3.5-Omni-Plus 63.3%
  15. 15 Nova 2.0 Pro Preview 61.6%
  16. 16 Granite 4.2 30B 61.4%
  17. 17 Nova 2 Lite 60.3%
  18. 18 Nova 2 Omni Preview 58.3%
  19. 19 Seed1.8 57.2%
  20. 20 LFM2.5-2.6B 56.9%
  21. 21 Nemotron 3 Nano 30B-A3B 53.8%
  22. 22 Nemotron-Cascade 2 30B-A3B 52.9%
  23. 23 LFM2.5-8B-A1B 48.5%
  24. 24 LFM2.5-VL-3B 32.5%
#ModelScoreSettingDateSource
1 Qwen3.7-Max 75% May 2026 Model card
2 Seed2.0 Pro 73.4% Feb 2026 Model card
3 Qwen3.5 397B-A17B 72.9% Feb 2026 Model card
4 Qwen3.7-Plus 72.9% Jun 2026 Model card
5 Qwen3.5-122B-A10B 72.2% Feb 2026 Model card
6 Spark-X2.6-4B 71.8% Oct 2026 Model card
7 LFM3-1.7B 71.4% Oct 2026 Model card
8 Ling-2.6-1T 70.6% Apr 2026 Model card
9 Trinity Large Thinking 70.1% Apr 2026 Model card
10 Qwen3-Max 67.7% Jan 2026 Model card
11 ERNIE 5.0 66.5% Feb 2026 Model card
12 Keye-VL-2.0-30B-A3B 65.7% May 2026 Model card
13 LFM2.5-24B-A2B 65% Oct 2026 Model card
14 Qwen3.5-Omni-Plus 63.3% Mar 2026 Model card
15 Nova 2.0 Pro Preview 61.6% Dec 2025 Model card
16 Granite 4.2 30B 61.4% Aug 2026 Model card
17 Nova 2 Lite 60.3% Dec 2025 Model card
18 Nova 2 Omni Preview 58.3% Dec 2025 Model card
19 Seed1.8 57.2% Dec 2025 Model card
20 LFM2.5-2.6B 56.9% Aug 2026 Model card
21 Nemotron 3 Nano 30B-A3B 53.8% Dec 2025 Model card
22 Nemotron-Cascade 2 30B-A3B 52.9% Mar 2026 Model card
23 LFM2.5-8B-A1B 48.5% May 2026 Model card
24 LFM2.5-VL-3B 32.5% Aug 2026 Model card

The benchmark's own leaderboard has 80 entries, 65 of them models we track. We plot every source in the charts below.

3 models have held number one since its release in Jul 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

OpenAI Anthropic Amazon Qwen ByteDance Seed
Own leaderboard Not tested since Dec 2025
Claude Opus 4.5 77.5%
Lab claims 4 leaders
Qwen3.7-Max 75%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Claude Opus 4.5 Claude Sonnet 4.5 o3 xLAM-2-32b-fc-r Cohere Command A Mistral Large 2 Palmyra X4 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Claude Opus 4.5 o3

4 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

xAI
Amazon
NVIDIA
ByteDance Seed
Qwen
Baidu
Arcee AI
Ant Group
Kuaishou
Liquid AI
IBM
iFlytek
DateLaunch postScoreAs the post writes it
27 Oct 2026 LFM3-1.7B 71.4 BFCL V4
22 Oct 2026 Spark-X2.6-4B 71.8 BFCL-V4
13 Oct 2026 LFM2.5-24B-A2B 65 BFCLv4
25 Aug 2026 Granite 4.2 30B 61.4 BFCL (v4)
12 Aug 2026 LFM2.5-VL-3B 32.5 BFCL V4
4 Aug 2026 LFM2.5-2.6B 56.9 BFCLv4
1 Jun 2026 Qwen3.7-Plus 72.9 BFCL-V4
28 May 2026 LFM2.5-8B-A1B 48.5 BFCLv4
25 May 2026 Keye-VL-2.0-30B-A3B 65.7 BFCL-V4
20 May 2026 Qwen3.7-Max 75 BFCL-V4
29 Apr 2026 Ling-2.6-1T 70.6 BFCL-V4
1 Apr 2026 Trinity Large Thinking 70.1 BCFLv4
30 Mar 2026 Qwen3.5-Omni-Plus 63.3 BFCL-V4
19 Mar 2026 Nemotron-Cascade 2 30B-A3B 52.9 BFCL v4
24 Feb 2026 Qwen3.5-122B-A10B 72.2 BFCL-V4
16 Feb 2026 Qwen3.5 397B-A17B 72.9 BFCL-V4
14 Feb 2026 Seed2.0 Pro 73.4 BFCL-v4
6 Feb 2026 ERNIE 5.0 66.5 BFCL v4
23 Jan 2026 Qwen3-Max 67.7 BFCL-V4
18 Dec 2025 Seed1.8 57.2 BFCL-v4
15 Dec 2025 Nemotron 3 Nano 30B-A3B 53.8 BFCL v4
2 Dec 2025 Nova 2 Omni Preview 58.3 BFCL v4 Overall FC
2 Dec 2025 Nova 2 Lite 60.3 BFCL v4 Overall FC
2 Dec 2025 Nova 2.0 Pro Preview 61.6 BFCL v4 Overall FC
19 Nov 2025 Grok 4.1 Fast Berkeley Function Calling v4 Benchmark, overall accuracy (chart)

11 system cards report it, from 7 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
13 Jun 2026 Ling-2.6-1T technical report 70.6 BFCL-v4, accuracy (Table 5) Capabilities
9 Jun 2026 Keye-VL-2.0-30B-A3B technical report 65.7 BFCL-V4 (Table 4) Capabilities
17 Apr 2026 Qwen3.5-Omni-Plus technical report 63.3 BFCL-V4 (Table 4) Capabilities
21 Mar 2026 Seed1.8 model card 57.2 BFCL-v4 (Table 4) Capabilities
19 Mar 2026 Nemotron-Cascade 2 30B-A3B technical report 52.9 BFCL v4, thinking mode, avg@1 (Table 1) Capabilities
14 Feb 2026 Seed2.0 Pro model card 73.4 BFCL-v4 (Table 11) Capabilities
4 Feb 2026 ERNIE 5.0 technical report 66.5 BFCL v4 (Table 2) Capabilities
23 Dec 2025 Nemotron 3 Nano 30B-A3B technical report 53.8 BFCL v4 (Table 3) Capabilities
2 Dec 2025 Nova 2 Omni Preview technical report 58.3 BFCL v4 Overall FC (Table 5) Capabilities
2 Dec 2025 Nova 2 Lite technical report 60.3 BFCL v4 Overall FC (Table 2) Capabilities
2 Dec 2025 Nova 2.0 Pro Preview technical report 61.6 BFCL v4 Overall FC (Table 2) Capabilities

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic Z.ai OpenAI Meta
BFCL V4 7 leaders
Claude Opus 4.5 77.5%
Berkeley Function Calling Leaderboard (BFCL) Passed 85% in Jul 2024, by Llama 3.1-405B
Llama 3.1-405B 88.5%
BFCL V4 3 leaders
Claude Opus 4.5 77.5%
Replaces Berkeley Function Calling Leaderboard (BFCL)

More in Tool use

All 125 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Muse Spark 1.1 88.1%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Claw-Eval In use 300 human-verified agent tasks (service orchestration, multimodal, multi-turn dialogue) audited via traces and snapshots. Rubric scores with Pass^3 consistency. Claude Sonnet 4.6 81.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
  • MCPMark In use Agent stress-tests realistic MCP tool use (GitHub, Notion, Postgres, browser). Graded by programmatic verification scripts. Kimi K3 94.5%