Qwen3 Max
Flagship Qwen3 model for coding agents, complex reasoning, and tool use
Shogo Agent Index
Estimated#79 of 9225.7range 16.2–35.3 · 11% of weight measured
Agentic coding
27
estimated
Long-horizon autonomy
33
30% measured
Knowledge-work agents
23
estimated
Computer use
15
estimated
Security & context learning
27
50% measured
Price
- Input
- $0.36/M
- Output
- $1.43/M
- Cache read
- $0.072/M
- 3:1 blend
- $0.63/M
- Agent-effective
- $0.18/M
Agent-effective = 20 input tokens per output token, 85% cache hits. Why.
Agentic benchmarks
| Benchmark | Score | Normalised | Index weight | Harness / source |
|---|---|---|---|---|
| SWE-bench Verified | est. | 8% | predicted 62 | |
| Terminal-Bench 2.0 | est. | 8% | predicted 24 | |
| DeepSWE | est. | 6% | predicted 19 | |
| FrontierSWE | est. | 5% | predicted 3 | |
| CursorBench | est. | 4% | predicted 24 | |
| GSO-Bench | est. | 4% | predicted 6 | |
| METR Time Horizons | est. | 10% | predicted 60 | |
| Vending-Bench 2 | $72 | 6% | source | |
| PostTrainBench | est. | 4% | predicted 15 | |
| APEX-Agents | est. | 8% | predicted 20 | |
| GDPval | est. | 7% | predicted 26 | |
| Remote Labor Index | est. | 5% | predicted 1 | |
| DeepResearch Bench | est. | 5% | predicted 44 | |
| OSWorld 2.0 | est. | 6% | predicted 2 | |
| OSWorld | est. | 4% | predicted 35 | |
| Cybench | est. | 5% | predicted 39 | |
| CL-bench | 14.5% | 5% | source |
Other benchmarks
| Benchmark | Category | Score | Source |
|---|---|---|---|
| DTBench | Knowledge | 82.1% | source |
| Fiction.LiveBench | Long context | 66.7% | source |
| FrontierMath Tiers 1-3 | Math | 18.9% | source |
| GPQA Diamond | Reasoning | 72.6% | source |
| LMCA | Knowledge | 28.3% | source |
| MATH Level 5 | Math | 97.1% | source |
| OTIS Mock AIME | Math | 73.3% | source |
| SimpleQA Verified | Knowledge | 48.7% | source |