Berkeley Function-Calling Leaderboard V4
42 models with a score. Leader: Opus 4.5 with 77.47 %.
| Benchmark profile | |
|---|---|
| Task | allgemein - Werkzeug- und Funktionsaufrufe, agentische Ablaeufe |
| Metric | Overall Acc ueber Non-Live, Live, Multi-Turn, Web Search, Memory und Relevanzerkennung |
| Direction | ↑ higher is better |
| Limitation | Naechste Verwandte der Strukturausgabe (Argumente nach Schema), misst aber Aufrufe, nicht das Befuellen eines festen JSON-Objekts. FC- und Prompt-Eintraege desselben Modells unterscheiden sich deutlich; der Katalog nimmt FC, wo vorhanden. Stand April 2026 - neuere Modelle (Qwen3.5/3.6, Gemma 4, DeepSeek V4) fehlen. |
| As of | 12 Apr 2026 |
| Source | gorilla.cs.berkeley.edu |
| Licence | Apache-2.0 — Berkeley Function Calling Leaderboard (Gorilla) |
Ranking (42)
No entry matches the filter.
| Model | Rank | Score | Developer | EU processing: InputEU in | EU processing: OutputEU out | all providers: Inputall in | all providers: Outputall out | Open weights | GPU memory (4-bit, derived)GPU memory |
|---|---|---|---|---|---|---|---|---|---|
| Opus 4.5 | 1 | 77.47 % | Anthropic | €4.77 | €23.85 | €2.24 | €11.20 | – no | – |
| Sonnet 4.5 | 2 | 73.24 % | Anthropic | €2.68 | €13.42 | €1.34 | €6.72 | – no | – |
| GLM 4.6 | 3 | 72.38 % | Z.ai | – | – | €0.39 | €1.57 | ✓ yes | from 201 GB |
| Grok 4.1 Fast Reasoning | 4 | 69.57 % | xAI | – | – | €0.18 | €0.45 | – no | – |
| Claude Haiku 4.5 | 5 | 68.7 % | Anthropic | €0.89 | €4.47 | €0.45 | €2.24 | – no | – |
| o3 | 6 | 63.05 % | OpenAI | – | – | €0.90 | €3.58 | – no | – |
| o3 2025-04-16 | 6 | 63.05 % | OpenAI | €0.99 | €3.94 | €0.90 | €3.58 | – no | – |
| Kimi K2 Instruct | 8 | 59.06 % | Moonshot AI | – | – | €0.51 | €2.06 | ✓ yes | from 554 GB |
| Grok 4.1 Fast Non-Reasoning | 9 | 58.29 % | xAI | – | – | €0.18 | €0.45 | – no | – |
| Gemini 2.5 Flash | 10 | 56.2 % | €0.27 | €2.24 | €0.13 | €1.12 | – no | – | |
| GPT-5.2 | 11 | 55.87 % | OpenAI | €1.72 | €13.79 | €0.78 | €6.27 | – no | – |
| GPT-5 mini | 12 | 55.46 % | OpenAI | €0.25 | €1.97 | €0.11 | €0.90 | – no | – |
| GPT-4.1 | 13 | 54 % | OpenAI | €1.97 | €7.87 | €0.90 | €3.58 | – no | – |
| GPT-4.1-2025-04-14 | 14 | 53.96 % | OpenAI | €0.99 | €3.94 | €0.90 | €3.58 | – no | – |
| o4-mini 2025-04-16 | 15 | 53.24 % | OpenAI | €0.54 | €2.17 | €0.49 | €1.97 | – no | – |
| Qwen3 235B A22B Instruct 2507 | 16 | 52.15 % | qwen-alibaba | €0.08 | €0.53 | €0.099 | €0.39 | ✓ yes | from 134 GB |
| GPT-5 nano | 17 | 51.45 % | OpenAI | €0.049 | €0.39 | €0.022 | €0.18 | – no | – |
| gpt-4.1-mini | 18 | 50.45 % | OpenAI | €0.39 | €1.53 | €0.18 | €0.72 | – no | – |
| GPT-4.1-mini-2025-04-14 | 18 | 50.45 % | OpenAI | €0.20 | €0.79 | €0.18 | €0.72 | – no | – |
| Qwen3-32B | 20 | 48.7 % | qwen-alibaba | €0.28 | €0.28 | €0.072 | €0.25 | ✓ yes | from 23 GB |
| Qwen3 8B | 21 | 42.57 % | Alibaba Cloud | – | – | – | – | ✓ yes | from 9 GB |
| Qwen3 30B A3B Instruct 2507 | 22 | 41.39 % | Alibaba Cloud | €0.089 | €0.27 | €0.089 | €0.27 | ✓ yes | from 21 GB |
| Qwen3-14B | 23 | 41 % | Alibaba Cloud | €0.18 | €0.90 | €0.11 | €0.21 | ✓ yes | from 13 GB |
| Mistral Large 2.1 | 24 | 38.4 % | Mistral AI | – | – | – | – | ✓ yes | from 72 GB |
| Llama 4 Maverick | 25 | 37.29 % | Meta | – | – | €0.16 | €0.52 | ✓ yes | from 225 GB |
| Mistral Small 3.2 24B | 26 | 37.1 % | Mistral AI | €0.061 | €0.16 | €0.061 | €0.16 | ✓ yes | from 18 GB |
| Gemini 2.5 Flash-Lite | 27 | 36.9 % | – | – | €0.045 | €0.18 | – no | – | |
| gpt-4.1-nano | 28 | 33.05 % | OpenAI | €0.10 | €0.39 | €0.10 | €0.39 | – no | – |
| GPT-4.1-nano-2025-04-14 | 28 | 33.05 % | OpenAI | €0.049 | €0.20 | €0.045 | €0.18 | – no | – |
| Llama 3.3 70B | 30 | 31.9 % | Meta | €0.45 | €0.65 | €0.09 | €0.29 | ✓ yes | from 43 GB |
| Gemma 3 12B | 31 | 30.4 % | – | – | €0.045 | €0.13 | ✓ yes | from 11 GB | |
| Gemma 3 27B | 32 | 29.5 % | €0.089 | €0.27 | €0.072 | €0.14 | ✓ yes | from 20 GB | |
| Phi-4 | 33 | 28.8 % | Microsoft | – | – | €0.063 | €0.13 | ✓ yes | from 13 GB |
| Llama 4 Scout 17B-16E | 34 | 28.13 % | Meta | – | – | €0.09 | €0.27 | ✓ yes | from 64 GB |
| Mistral Nemo Instruct | 35 | 27.63 % | Mistral AI | €0.016 | €0.025 | €0.016 | €0.025 | ✓ yes | from 11 GB |
| Nova 2 Lite | 36 | 27.1 % | Amazon | €0.34 | €2.82 | €0.15 | €1.23 | – no | – |
| Llama 3.1 8B Instruct | 37 | 25.83 % | Meta | €0.016 | €0.041 | €0.018 | €0.036 | ✓ yes | from 9 GB |
| Nova Pro | 38 | 24.97 % | Amazon | €0.82 | €3.29 | €0.36 | €1.43 | – no | – |
| Nova Micro | 39 | 22.29 % | Amazon | €0.036 | €0.14 | €0.031 | €0.13 | – no | – |
| Llama 3.2 3B | 40 | 21.95 % | Meta | – | – | €0.046 | €0.30 | ✓ yes | from 6 GB |
| Gemma 3 4B | 41 | 19.62 % | – | – | €0.018 | €0.036 | ✓ yes | from 7 GB | |
| Llama 3.2 1B | 42 | 10.82 % | Meta | – | – | €0.024 | €0.18 | ✓ yes | from 5 GB |
Equal scores share a rank. When filtering, the rank stays the position in the whole list. GPU memory “from …” is a lower bound (parameters × 0.55 GB + 4 GB). Prices: € per 1M tokens. Source: providers' pricing pages, read and checked by searchit. As of 7 Nov 2026.
Price and performance
Scores: Berkeley Function-Calling Leaderboard V4, Apache-2.0 — Berkeley Function Calling Leaderboard (Gorilla). Taken unchanged from the source.