Models › Qwen 3.8 Max
Alibaba • Launched August 2026, updated September 2026
Qwen 3.8 Max is Alibaba’s largest multimodal reasoning model yet
Starting provider API rate; context tiers apply
See what 3.8 Max does best, then review its limits, benchmarks, and pricing below
Qwen 3.8 Max is Alibaba’s largest multimodal reasoning model yet
Every 3.8 Max result below comes from a named source and is labeled by protocol, so each score reflects how it was measured.
| Benchmark | Score | Evaluation details | Source |
|---|---|---|---|
GPQA Diamond The highest-quality subset of Graduate-Level Google-Proof Q&A, designed to test expert-level scientific reasoning in biology, physics, and chemistry. |
92.8% Accuracy |
Version: GPQA Diamond 198 · Split: 198 Diamond questions · Harness: Artificial Analysis independent evaluation · Snapshot: 0902 build | Methodology Artificial Analysis Independently reproduced |
Humanity's Last Exam A 2,500-question expert-level benchmark spanning dozens of academic fields. Tool-assisted and no-tools results are separate protocols. |
43.1% Accuracy |
Version: Humanity's Last Exam · Harness: Artificial Analysis independent evaluation · Snapshot: 0902 build | Methodology Artificial Analysis Independently reproduced |
Terminal-Bench Hard A harder subset of Terminal-Bench focused on multi-step terminal tasks that trip up most agents. |
38.9% Task success rate |
Version: Terminal-Bench Hard · Harness: Artificial Analysis independent evaluation · Snapshot: 0902 build | Methodology Artificial Analysis Independently reproduced |
OSWorld-Verified A computer-use benchmark that scores whether an agent can complete real desktop tasks inside a verified operating-system environment. |
86.1% Task success rate |
Version: OSWorld-Verified · Harness: Alibaba internal evaluation · Reasoning effort: high | Methodology Alibaba (Qwen team) Provider-reported |
DeepSWE 1.1 A long-horizon software-engineering benchmark with 113 original tasks graded by hand-written tests. |
69.3% Pass@1 |
Version: DeepSWE 1.1 · Harness: Alibaba internal evaluation · Snapshot: 0902 build, up from 56.6% | Methodology Alibaba (Qwen team) Provider-reported |
Arena rankings capture how people vote between anonymous responses. 3.8 Max’s human preference standing is reported per leaderboard, since results can vary widely by task
.8 Max ranks first on Arena’s Code: WebDev leaderboard for building working web apps, scoring 1,691 points
Qwen 3.8 Max runs a 983,616-token context window at one flat rate, with image input and adjustable reasoning, priced and detailed below
| Model | |
| Provider | Alibaba (Qwen) |
| Release date | Sep 2, 2026 (0902 snapshot) |
| Catalog added | Sep 22, 2026 |
| Context window | 983,616 tokens |
| Maximum output | 131,072 tokens |
| Knowledge cutoff | Not available in reviewed sources |
| License | Proprietary |
| Capabilities and API | |
| Parameters | 2.4T total • ~95B active (MoE) |
| Input types | Image, Text |
| Output types | Text |
| Reasoning | Supported (thinking/non-thinking modes, adjustable effort) |
| Provider endpoint tool parameters | Accepts tool parameters |
| Base provider API input | $2.00 / 1M tokens |
| Base provider cached input | $0.25 / 1M tokens |
| Base provider API output | $6.00 / 1M tokens |
Provider token rates are context-dependent. The base rates apply below the listed prompt thresholds; the matching tier applies at or above each threshold. These are reference provider API rates, not Pyxa AI checkout charges — access in Pyxa AI follows the allowances of your Pyxa AI plan.
Each 3.8 Max benchmark and specification links to its original source, so you can check where every figure comes from
Open another model’s detail page or compare it directly with Qwen 3.8 Max. Suggestions come from the current catalog
Put Qwen 3.8 Max to work on your coding, research, and long-context tasks inside Pyxa AI