AgentWorldBench (Android subset) — Qwen-published agent benchmark w/ explicit per-domain subset naming; keep domain subsets separate
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6 | Anthropic | 61.7 | 1 | 2026-06-24 |
| 2 | Claude Opus 4.8 | Anthropic | 61.5 | 1 | 2026-06-24 |
| 3 | Gemini 3.1 Pro | 61.4 | 1 | 2026-06-24 | |
| 4 | Qwen AgentWorld 397B A17B | Alibaba | 60.2 | 1 | 2026-06-24 |
| 5 | GPT-5.4 | OpenAI | 60 | 1 | 2026-06-24 |
| 6 | GLM-5.1 | Z.ai | 59.1 | 1 | 2026-06-24 |
| 7 | Kimi K2.6 | Moonshot | 58.9 | 1 | 2026-06-24 |
| 8 | Qwen AgentWorld 35B A3B | Alibaba | 58.2 | 1 | 2026-06-24 |
| 9 | Claude Sonnet 4.6 | Anthropic | 58 | 1 | 2026-06-24 |
| 10 | Qwen3.6 Plus | Alibaba | 57.6 | 1 | 2026-06-24 |
| 11 | DeepSeek-V4-Pro | DeepSeek | 55.2 | 1 | 2026-06-24 |
| 12 | Qwen3.5 397B A17B | Alibaba | 54.9 | 1 | 2026-06-24 |
| 13 | Qwen3.5 35B A3B | Alibaba | 53.2 | 1 | 2026-06-24 |
| 14 | MiniMax M2.7 | MiniMax | 52.4 | 1 | 2026-06-24 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.