← Back to directory
Alibaba (Qwen)
Qwen3-Max
Released Sep 24, 2025 · knowledge cutoff 2025-06
- Status
- Superseded
- Location
- China
- Modality
- Multimodal
- Context window
- 262K
- Max output
- 66K
- Speed
- —
- Price ($/MTok in / out)
- $1.20 / $6
- Cost per task
- —
- Open weights
- No
Benchmarks5 of 8 reported
MMLU-Pro
85.4%
GPQA Diamond
85.4%
SWE-bench Verified
69.6%
AIME (latest)
92.3%
LMArena Elo
1428
Terminal-Bench 2.1
—
Humanity's Last Exam
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Reliable multi-step tool calling — well-formed calls with correct arguments
- Fast for its size thanks to sparse activation
- Chinese and English both first-class rather than one as an afterthought
Weaknesses
- Closed weights, unlike the rest of the Qwen line
- Long agentic sessions get verbose, and the verbosity is where the cost goes
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor