AutomationBench
An agent benchmark for completing automation workflows in reproducible task environments.
Top models on AutomationBench — October 29, 2026
As of October 29, 2026, MiniMax M3.1 Flash leads the AutomationBench leaderboard with 62.2% , followed by DeepSeek V4.1 Flash 1017 (60.5%) and Gemini 4 Carbon (53.9%).
MiniMax M3.1 Flash
MiniMax
DeepSeek V4.1 Flash 1017
DeepSeek
Gemini 4 Carbon
27 modelsAgentic6% of Agentic reference weightCurrentUpdated October 29, 2026
Leaderboard (27 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | MiniMax M3.1 FlashMiniMax | 62.2% | Not reported | Closed |
| 2 | DeepSeek V4.1 Flash 1017DeepSeek | 60.5% | Not reported | Open |
| 3 | Gemini 4 CarbonGoogle | 53.9% | Not reported | Closed |
| 4 | Atria Dawn PreviewShanghai Artificial Intelligence Laboratory | 53.8% | Not reported | Open |
| 5 | MiMo-V2.6-ProXiaomi | 53.1% | Not reported | Open |
| 6 | Ling 3.1 FlashInclusionAI | 52.5% | Not reported | Open |
| 7 | MiMo-V2.6-FlashXiaomi | 52.3% | Not reported | Open |
| 8 | Gemini 4 ArgonGoogle | 51.3% | Not reported | Closed |
| 9 | Ultramarine 1 FlashPaper Instruments | 50.2% | Not reported | Closed |
| 10 | Muse Spark 1.3Meta | 49.4% | Not reported | Closed |
| 11 | GLM-5.3-FlashZ.AI | 48.8% | Not reported | Open |
| 12 | GLM-5.3Z.AI | 48.2% | Not reported | Open |
| 13 | Step 5 PreviewStepFun | 44.0% | Not reported | Open |
| 14 | GPT-6 AstraOpenAI | 41.4% | Not reported | Closed |
| 15 | LFM3-1.7BLiquidAI | 41.3% | Not reported | Open |
| 16 | Claude Opus 5.5Anthropic | 40.0% | Not reported | Closed |
| 17 | GPT-6.1 SolOpenAI | 36.1% | Not reported | Closed |
| 18 | GPT-6 SolOpenAI | 33.2% | Not reported | Closed |
| 19 | Hy4 previewTencent | 32.1% | Not reported | Open |
| 20 | DeepSeek V4 Pro 0813DeepSeek | 31.8% | Not reported | Open |
| 21 | Claude Fable 5.1Anthropic | 31.4% | Not reported | Closed |
| 22 | Kimi K3Moonshot AI | 30.8% | Not reported | Open |
| 23 | Gemini 3.7 FlashGoogle | 30.4% | Not reported | Closed |
| 24 | LFM2.5-24B-A2BLiquidAI | 27.7% | Not reported | Open |
| 25 | Qwen3.8 MaxAlibaba | 27.3% | Not reported | Open |
| 26 | Claude Opus 5Anthropic | 26.0% | Not reported | Closed |
| 27 | DeepSeek V4 Flash 0731DeepSeek | 25.1% | Not reported | Open |
According to BenchLM.ai, MiniMax M3.1 Flash leads the AutomationBench benchmark with a score of 62.2%, followed by DeepSeek V4.1 Flash 1017 (60.5%) and Gemini 4 Carbon (53.9%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
27 models have been evaluated on AutomationBench. The benchmark falls in the Agentic category. BenchAlign v5.8 gives AutomationBench 6% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About AutomationBench
Year
2026
Tasks
600 public automation tasks
Format
Agent task-completion score
Difficulty
Long-horizon automation
Moonshot evaluates the 600-task public subset while following the official GitHub setup. BenchLM stores this provider-run result separately from Artificial Analysis AutomationBench-AA.
Freshness and provenance
Version
AutomationBench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does AutomationBench measure?
An agent benchmark for completing automation workflows in reproducible task environments.
Which model scores highest on AutomationBench?
MiniMax M3.1 Flash by MiniMax currently leads with a score of 62.2% on AutomationBench.
How many models are evaluated on AutomationBench?
27 AI models have published results on AutomationBench in the BenchLM catalog.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.