Alibaba · 2026-02-20
Qwen3-Max
256k context · reasoning · multimodal · tools · closed
Capability
81.6
Coding
80.6
Agents
62.1
Science
83.1
TTFT
260 ms
p50, US-East, streaming
Output speed
92 tok/s
p95 latency
10.2 s
Availability
99.70%
Cost/task
$0.026
Cost/correct task
$0.036
30-day regressions
0
Measured runs
11,220
Last tested 22 min ago
LIVE_MEASUREDEWR 2.56 · human_io 0
TTFT spark · primary endpoint
Category scores
- Math85.4
- Reasoning82.8
- Coding80.6
- Agents62.1
- Multimodal78.4
- Science83.1
- Long context85.6
- Private tasks66.4
Serving stacks
- TogetherLIVE_MEASURED260 ms · 92 tok/s
Reliability on Together: invalid JSON 1.8% · broken tools 0.7% · replay 91.2%
| Benchmark | Score | $/task | Median s | Evidence |
|---|---|---|---|---|
| GPQA Diamond public · n=198 | 78.2 | $0.0056 | 9.7 | LIVE_MEASURED |
| SWE-bench Verified public · n=500 | 69.4 | $0.0070 | 9.4 | LIVE_MEASURED |
| Terminal Bench public · n=80 | 40.8 | $0.0070 | 27.7 | LIVE_MEASURED |
| OSWorld public · n=369 | 32.1 | $0.0070 | 27.0 | LIVE_MEASURED |
| FrontierMath public · n=300 | 29.6 | $0.0056 | 20.1 | LIVE_MEASURED |
| LiveCodeBench public · n=612 | 79.4 | $0.0070 | 9.7 | LIVE_MEASURED |
| MMMU public · n=1730 | 75.2 | $0.0056 | 9.6 | LIVE_MEASURED |
| LongContext-128k public · n=120 | 85.6 | $0.0056 | 9.9 | LIVE_MEASURED |
| GBU Private Rotating — Sep 2026 private · n=240 | 66.4 | $0.0056 | 9.4 | LIVE_MEASURED |
| Agent Episodes public · n=96 | 62.1 | $0.0070 | 29.6 | LIVE_MEASURED |
| Structured JSON public · n=400 | 91.1 | $0.0056 | 10.0 | LIVE_MEASURED |
| CAD Science Lab public · n=64 | 63.8 | $0.0056 | 9.3 | LIVE_MEASURED |