Benchmarks / Agentic
AutomationBench (Zapier)
Zapier's agent benchmark (v1.0.6): end-to-end business workflows across 47 real tools in six functions, graded strictly on the final environment state (every assertion must pass). Private held-out set.
AutomationBench (Zapier)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Gemini 4 | high | 51.3% |
| 2 | Sonnet 5.5 | max | 44.8% |
| 3 | Opus 5.5 | max | 42.5% |
| 4 | GPT-6 Astra | max | 41.4% |
| 5 | GPT-6 Sol | xhigh | 33.2% |
| 6 | Gemini 3.7 Flash | high | 30.4% |
| 7 | Gemini 3.8 Flash | medium | 29.7% |
| 8 | GPT-5.6 Sol | max | 28.8% |
| 9 | Opus 5 | max | 26.9% |
| 10 | GPT-5.6 Terra | max | 23.6% |
| 11 | Kimi K3 | max | 22.7% |
| 12 | Fable 5.1 | max | 22.4% |
| 13 | GPT-6 Luna | max | 20.7% |
| 14 | Muse Spark 1.3 | max | 20.7% |
| 15 | DeepSeek V4 Flash | max | 18.1% |
| 16 | GPT-5.6 Luna | max | 17.1% |
| 17 | Gemini 3.6 Flash | high | 17.1% |
| 18 | Fable 5.0 | max | 17.1% |
| 19 | Opus 4.8 | xhigh | 16.9% |
| 20 | GPT-5.5 | xhigh | 16.6% |
| 21 | Gemini 3.5 Flash | high | 13.4% |
| 22 | Opus 4.7 | max | 13.4% |
| 23 | Haiku 5.5 | medium | 12.0% |
| 24 | GPT-5.4 | xhigh | 11.6% |
| 25 | Sonnet 5 | max | 10.7% |
| 26 | Gemini 3.1 Pro | high | 8.7% |
| 27 | GLM 5.2 | max | 7.8% |
| 28 | GLM 5.1 | max | 6.2% |
| 29 | Sonnet 4.6 | max | 5.3% |
| 30 | MiniMax M3 | max | 4.9% |
| 31 | Qwen 3.6 | high | 4.3% |
| 32 | Gemini 3.5 Flash-Lite | medium | 4.0% |
| 33 | Qwen 3.7 | – | 3.7% |
| 34 | Kimi K2.7 Code | max | 3.2% |
| 35 | Kimi K2.6 | max | 2.6% |
| 36 | Gemma 4 31B | max | 1.7% |
| 37 | Haiku 4.5 | xhigh | 0.5% |
| 38 | MiniMax M2.7 | high | 0.5% |
| 39 | gpt-oss-120b | high | 0.1% |