Head-to-head · Updated Nov 3, 2026
Qwen vs DeepSeek
The highest-scoring Qwen and DeepSeek models, Qwen3.8 Max and DeepSeek V4.1 Flash 1017, on independent tests for professional work, cost and speed.
Short answer
Qwen3.8 Max leads on 7 of 12 quality scores, including the overall Intelligence Index.
- FinanceQwen3.8 Max45.0 vs 42.2
- Legal workQwen3.8 Max41.0 vs 37.7
- HealthcareDeepSeek V4.1 Flash 101741.3 vs 40.8
- EngineeringQwen3.8 Max39.2 vs 37.5
- Strategy & operationsDeepSeek V4.1 Flash 101749.2 vs 47.3
- Coding agentsQwen3.8 Max38.9% vs 33.8%
- Expert knowledgeDeepSeek V4.1 Flash 101743.3% vs 43.1%
- Fewer made-up answersQwen3.8 Max27.1% vs 64.1%
- Lower costDeepSeek V4.1 Flash 1017$521 vs $6,354
- SpeedDeepSeek V4.1 Flash 1017174 tok/s vs 38 tok/s
Scores, cost and speed
| Measure | Qwen3.8 Max (0902) | DeepSeek V4.1 Flash 1017 (Max) |
|---|---|---|
| Overall | ||
| Intelligence IndexCombined score across nine evaluations | Leads: 33.6 | 32.5 |
| Professional work | ||
| Finance & accounting | Leads: 45.0 | 42.2 |
| Legal | Leads: 41.0 | 37.7 |
| Healthcare & medical | 40.8 | Leads: 41.3 |
| Engineering | Leads: 39.2 | 37.5 |
| Strategy & operations | 47.3 | Leads: 49.2 |
| Economics | Leads: 44.2 | 40.3 |
| Specific tests | ||
| Terminal-Bench 4.0Agentic coding tasks | Leads: 38.9% | 33.8% |
| FrontierCode 2.0Hard programming problems | 23.4% | Leads: 28.5% |
| Humanity’s Last ExamExpert-level questions | 43.1% | Leads: 43.3% |
| HealthBenchHealth conversations graded by physicians’ rubrics | — | — |
| GDPval-AAKnowledge-work deliverables, Elo | 1674 | Leads: 1746 |
| Hallucination rateAA-Omniscience; lower is better | Leads: 27.1% | 64.1% |
| Cost & speed | ||
| Input priceUS$ per 1M tokens | $2 | Leads: $0.30 |
| Output priceUS$ per 1M tokens | $6 | Leads: $1.20 |
| Cost to run the indexUS$, includes reasoning tokens | $6,354 | Leads: $521 |
| Output speedMedian tokens per second | 38 tok/s | Leads: 174 tok/s |
| Context windowTokens | 984K | Leads: 1M |
Scores, prices and speeds: Artificial Analysis, independently measured, checked Nov 3, 2026. Model names include the tested reasoning setting. Stubbi has not rerun these evaluations. Consumer apps such as ChatGPT, Claude and Gemini may use other models and add features like search and memory.
All measured Qwen and DeepSeek models
| Model and setting | Released | Index | Price in / out |
|---|---|---|---|
| Qwen3.8 Max (0902)Alibaba | Sep 2026 | 33.6 | $2 / $6 |
| DeepSeek V4.1 Flash 1017 (Max)DeepSeek | Oct 2026 | 32.5 | $0.30 / $1.20 |
| Qwen3.8 27B (Xhigh)Alibaba | Aug 2026 | 20.1 | $0.50 / $3 |
Test them on your own work
Public scores can’t show how each model handles your documents, clients or standards. Tell us the task, and we’ll consider a side-by-side comparison with the outputs you can judge.