Qwen4 Max is Alibaba's AI model, released 16/11/2026. Alibaba's launch post compares Qwen4-Max with Qwen3.8-Max-0902, Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra and Kimi K3.1 across 16 benchmarks. It leads on DeepSWE v1.1 (75.7), Agents' Last Exam v1.1, HLAB-Vals legal work and BioMysteryBench, and trails Claude Opus 5.5 and Kimi K3.1 clearly on Terminal-Bench 4.0 and ProgramBench. The launch page is script-rendered and could not be fetched, so every row here comes from the site article that reads Alibaba's chart; some chart rows are third-party scores cited by Alibaba.
Visit Qwen4 MaxAI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Qwen4-Max score on independent benchmarks?
Qwen4-Max ranks #13 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 42.9, as measured on 22 November 2026.
Where does Qwen4-Max rank against other models?
How does Qwen4-Max do on individual tests?
| Evaluation (run by Artificial Analysis) | Qwen4-Max | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 48.5% | 67.2% |
| FrontierCode 2.0 | 26.5% | 48.8% |
| Humanity's Last Exam | n/a | 74.3% |
| GDPval-AA (Elo) | 1809 | 1866 |
How has Qwen4-Max's score changed over time?
We have tracked Qwen4-Max on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Qwen4-Max system card report?
Alibaba's launch post compares Qwen4-Max with Qwen3.8-Max-0902, Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra and Kimi K3.1 across 16 benchmarks. It leads on DeepSWE v1.1 (75.7), Agents' Last Exam v1.1, HLAB-Vals legal work and BioMysteryBench, and trails Claude Opus 5.5 and Kimi K3.1 clearly on Terminal-Bench 4.0 and ProgramBench. The launch page is script-rendered and could not be fetched, so every row here comes from the site article that reads Alibaba's chart; some chart rows are third-party scores cited by Alibaba.
These are Alibaba (Qwen)'s own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Source: Qwen4-Max: Your New Teammate Has Arrived (16/11/2026). Last checked 22/11/2026.
Coding
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| DeepSWE v1.1Agentic coding | 75.7% | 69.3% | 74.2% | 75.2% |
| CodeArenaWeb development | 1750 Elo | 1676 | 1816 | 1742 |
| Terminal-Bench 4.0Effort level and harness not stated in the launch post | 48.5% | 38.9% | 59.6% | 56.1% |
| ProgramBench | 31.5% | 28% | 65% | 52.5% |
Agents & tool use
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| AA-Briefcase v1.1Score cited from Artificial Analysis | 1686 Elo | 1605 | 1801 | 1555 |
| JobBench | 68.5% | 65.3% | 75.4% | 66.2% |
| APEX-Agents v1.1Score cited from Mercor | 72.1% | 63.3% | 73.5% | 60% |
| HLAB-Vals-1.1 (legal)Score cited from the Vals Index | 21.3% | 15.4% | 9.2% | 6.3% |
| AutomationBench-AAScore cited from Artificial Analysis | 72.5% | 56.2% | 69.5% | 64.9% |
| Agents' Last Exam v1.1Separate panel in the launch chart; Qwen3.8-Max 27.9, Opus 5.5 43.4, Kimi K3.1 43.4 | 46.7% | 27.9% | 43.4% | n/a |
| AgentArena (net improvement)Separate panel in the launch chart; Opus 5.5 18.2 | 7.5 score | n/a | 18.2 | n/a |
| Work360Alibaba-designed benchmark (26 domains, 182 subdomains, 721 fields); useful for comparing Qwen generations only | 52.69 score | 47.74 | n/a | n/a |
Knowledge & science
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| BioMysteryBench | 85% | 84.1% | 81.9% | 84.4% |
| HealthBench Professional | 60% | 39.6% | 59.9% | 65.1% |
| Humanity's Last Exam | 51.7% | 43.9% | 59.8% | 54% |
Multimodal
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| ChartographyChart understanding | 54.2% | 42% | 66.3% | 67.8% |
| Video-IndexVideo understanding | 68.1% | 63.2% | 74.8% | 73.8% |
Safety & honesty
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| Last Secure Code Bench (function level, correct and secure)Share of functions both correct and secure | 31.8% | 22% | n/a | n/a |
| Last Secure Code Bench (function, file and repository level)Share secure at all three levels | 14.7% | 8% | n/a | n/a |
Other
| Benchmark | Qwen4-Max | Qwen3.8-Max | Claude Opus 5.5 | GPT-6.1 Sol |
|---|---|---|---|---|
| Base model: Knowledge (10 x bits per byte)Qwen4-Max-Base; lower is better; Alibaba in-house evaluation. Kimi-K3.1-Pro is a post-trained model so not like for like. GLM-5.3 is 5.87 | 4.86 BPB x10 | 5.6 | n/a | n/a |
| Base model: Code (10 x bits per byte)Qwen4-Max-Base; lower is better; Alibaba in-house evaluation. GLM-5.3 is 1.90 | 1.67 BPB x10 | 1.88 | n/a | n/a |
| Base model: QA (10 x bits per byte)Qwen4-Max-Base; lower is better; Alibaba in-house evaluation. GLM-5.3 is 3.20 | 2.94 BPB x10 | 3.1 | n/a | n/a |
| Base model: STEM (10 x bits per byte)Qwen4-Max-Base; lower is better; Alibaba in-house evaluation. GLM-5.3 is 3.24 | 2.7 BPB x10 | 3.04 | n/a | n/a |
What did safety testing find for Qwen4-Max?
Safety level: No formal system card; vendor safety section in launch post
- Last Secure Code Bench: share of functions both correct and secure rose from 22.0% to 31.8%; secure at function, file and repository level rose from 8% to 14.7%.
- Alibaba says internal experts reviewed long agent runs, including the full chain of thought, and found no capabilities posing large-scale cybersecurity risks.
- Alibaba says internal experts found no significant CBRN uplift risk, and that the model did not over-refuse.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our Qwen4-Max deep dive.
Alibaba Model Timeline
1M tokens context
1M tokens context
131k tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
