Qwen 3.8 Max and DeepSeek V4 Pro are the two strongest Chinese coding models available. Alibaba’s flagship has 2.4T parameters with multimodal capabilities. DeepSeek’s champion has 1.6T parameters with the SWE-bench Verified record (80.6%).
The choice comes down to whether you need multimodal capabilities or proven SWE-bench performance.
Specs comparison
| Spec | Qwen 3.8 Max | DeepSeek V4 Pro |
|---|---|---|
| Total params | 2.4T | 1.6T MoE |
| Active params | 95B | 49B |
| Architecture | Sparse MoE + hybrid attention | MoE |
| Context window | 1M | 1M |
| Multimodal | Yes (text + vision) | No (text only) |
| License | TBD (open weights next week) | MIT |
| Open weights | Next week | Available now |
| API price | Not published (~$3-5/$10-15 est.) | $2.19/$8.76 |
Benchmarks
| Benchmark | Qwen 3.8 Max | DeepSeek V4 Pro |
|---|---|---|
| Text Arena | #5 | not published |
| Vision Arena | #2 | N/A (text only) |
| Frontend Code Arena | #4 | not published |
| SWE-Bench Verified | not published | 80.6% (record) |
| Codeforces | not published | 3206 |
| LiveCodeBench | not published | 93.5% |
| Autonomous coding | 16 days | not published |
DeepSeek V4 Pro has the stronger published coding benchmarks. Qwen 3.8 Max has the stronger Arena rankings and autonomous coding evidence.
The MIT license advantage
DeepSeek V4 Pro: MIT license. The most permissive license available. You can use it for anything, modify it, redistribute it, and embed it in commercial products. No restrictions.
Qwen 3.8 Max: License TBD. Open weights coming next week, but the license has not been confirmed. Alibaba has historically used Apache 2.0 for Qwen models, but this model is larger and may have different terms.
If licensing matters (enterprise use, commercial products, fine-tuning), DeepSeek V4 Pro’s MIT license is the safest choice today.
Pricing
DeepSeek V4 Pro: $2.19/$8.76 per 1M tokens. This is the confirmed price.
Qwen 3.8 Max: Not published. Expected $3-$5/$10-$15 based on Qwen’s pricing history. If priced at the lower end, it will be competitive with DeepSeek. If priced at the higher end, DeepSeek is cheaper.
For budget-conscious developers, DeepSeek V4 Pro is the known quantity. Qwen 3.8 Max pricing could be better or worse.
Multimodal capabilities
Qwen 3.8 Max: Full multimodal support. Vision Arena #2 ranking. Processes documents, images, and video natively.
DeepSeek V4 Pro: Text only. No vision capabilities.
If you need multimodal processing, Qwen 3.8 Max is the only option. For text-only coding, both are competitive.
Self-hosting
DeepSeek V4 Pro: Weights available now. MIT license. You can self-host, fine-tune, and modify today.
Qwen 3.8 Max: Weights coming next week. License TBD. Self-hosting will require multiple 80GB+ GPUs (2.4T total params).
If you need to self-host today, DeepSeek V4 Pro is the only option. If you can wait a week, Qwen 3.8 Max will be available.
When to use Qwen 3.8 Max
- You need multimodal capabilities (documents, images, video)
- You want 16-day autonomous coding capability
- You prefer Alibaba’s ecosystem
- You can wait a week for open weights
- You need Vision Arena-class performance
When to use DeepSeek V4 Pro
- You need the SWE-bench Verified record (80.6%)
- You want MIT license (most permissive)
- You need open weights today
- You want confirmed pricing ($2.19/$8.76)
- You only process text (no vision needed)
My take
Both are excellent coding models. The choice depends on three factors: benchmarks, licensing, and modality.
For proven coding benchmarks, DeepSeek V4 Pro wins. The 80.6% SWE-bench Verified record is the strongest published coding benchmark among open models. The Codeforces 3206 score and 93.5% LiveCodeBench are also impressive.
For multimodal coding, Qwen 3.8 Max wins. If you process documents, images, or video alongside code, Qwen 3.8 Max is the only option. The Vision Arena #2 ranking is strong.
For licensing, DeepSeek V4 Pro wins today. The MIT license is the most permissive available. Qwen 3.8 Max’s license is TBD.
My recommendation: if you only need text-based coding and want the best proven benchmarks, use DeepSeek V4 Pro. If you need multimodal capabilities or want to wait for the Qwen 3.8 Max open weights, hold off for a week and benchmark both.
For budget-conscious developers, neither is the cheapest option. GPT-5.6 Luna at $0.20/$1.20 and DeepSeek V4 Flash at $0.14/$0.28 are 10x cheaper for most tasks.
FAQ
Which is more powerful, Qwen 3.8 Max or DeepSeek V4 Pro?
They are competitive on different metrics. Qwen 3.8 Max leads on Arena rankings (#5 Text, #2 Vision, #4 Frontend Code). DeepSeek V4 Pro leads on SWE-bench Verified (80.6%). Direct comparison is limited.
Which is cheaper?
DeepSeek V4 Pro costs $2.19/$8.76 per 1M tokens. Qwen 3.8 Max pricing has not been published. If Qwen is priced at $3/$10, DeepSeek is cheaper. If Qwen is priced at $5/$15, DeepSeek is significantly cheaper.
Can I self-host both?
DeepSeek V4 Pro weights are available now (MIT license). Qwen 3.8 Max weights are coming next week. If you need self-hosting today, DeepSeek is the only option.
Does DeepSeek V4 Pro have vision capabilities?
No. DeepSeek V4 Pro is text-only. If you need multimodal processing, use Qwen 3.8 Max.
Which has better autonomous coding?
Qwen 3.8 Max demonstrated 16-day autonomous coding. DeepSeek V4 Pro has no comparable demonstration, though its SWE-bench score (80.6%) suggests strong coding capability.
Should I wait for Qwen 3.8 Max weights or use DeepSeek V4 Pro now?
If you need open weights today, use DeepSeek V4 Pro. If you can wait a week, benchmark both when Qwen 3.8 Max weights drop. The competition will benefit you either way.