DeepSeek V4 Pro 0813 vs Qwen3.8 Max (0902)
A side-by-side of live API prices, providers, speed, usage and benchmarks. No verdict: the right choice depends on your workload.
DeepSeek V4 Pro 0813 costs $0.1333 per 1M input and $0.40 per 1M output tokens at the standard tier, via Baidu (fp8); Qwen3.8 Max (0902) costs $2.00 and $6.00, via Alibaba. At a 3:1 input:output mix, DeepSeek V4 Pro 0813 is 93% cheaper than Qwen3.8 Max (0902).
Why these two are compared
The large mixture-of-experts models of DeepSeek (V4 Pro, GA release 0813) and Alibaba’s Qwen team (Qwen3.8 Max, snapshot 0902).
Side by side
| Measure | DeepSeek V4 Pro 0813 | Qwen3.8 Max (0902) |
|---|---|---|
| Developer | DeepSeek | Qwen (Alibaba) |
| Released | 2026-08-12 | 2026-09-03 |
| Standard price, input / output per 1M | $0.1333 / $0.40 | $2.00 / $6.00 |
| Blended price (3:1) | $0.20 | $3.00 |
| Cheapest standard provider | Baidufp8 | Alibaba |
| Cache read per 1M | $0.004242 | $0.25 |
| Providers at the standard tier | 21 | 1 |
| Context window (tokens) | 1,048,576 | 1,000,000 |
| Max output (tokens) | 943,718 | 131,072 |
| Input modalities | text | text, image, video |
| Long-prompt pricing | No threshold listed | No threshold listed |
| Fastest endpoint | 110 tok/sTogether AI | 36 tok/sAlibaba |
| Lowest latency | 466 msWafer (mxfp4) | 2.80 sAlibaba (us-east-1 region) |
| OpenRouter tokens (2026-10-16) | 161.08B | Not in named rows |
| AA intelligence | 36.0 | 45.4 |
| AA coding | 68.8 | 76.2 |
| AA agentic | 41.3 | 56.0 |
“Not reported” means the source publishes no value. AA scores are the Artificial Analysis indexes reported through OpenRouter (CC BY 4.0), not SF·SIX measurements.
Cost for three example workloads
| Workload | DeepSeek V4 Pro 0813 / month | Qwen3.8 Max (0902) / month |
|---|---|---|
| Support chatbot50,000 requests × 1,500 in / 400 out tokens | $18.00 | $270.00 |
| RAG / research assistant20,000 requests × 8,000 in / 700 out tokens | $26.93 | $404.00 |
| Batch summarisation10,000 requests × 10,000 in / 500 out tokens | $15.33 | $230.00 |
Standard headline prices; caching and long-context surcharges not included. Price your own workload.
DeepSeek V4 Pro 0813 providers
| Provider and terms | Input / 1M | Output / 1M | Throughput | Latency |
|---|---|---|---|---|
| Baidufp8 | $0.1333 | $0.40 | 42 tok/s | 1.42 s |
| Sail Researchfp4 | $0.448 | $1.16 | Not reported | Not reported |
| Wafermxfp4 | $0.25 | $2.50 | 96 tok/s | 466 ms |
| DeepSeek | $0.66 | $1.98 | 32 tok/s | 1.11 s |
| StreamLake | $0.66 | $1.98 | 45 tok/s | 2.20 s |
| Ionstream | $0.37 | $2.93 | 61 tok/s | 739 ms |
| Phala | $0.957 | $2.88 | 49 tok/s | 2.91 s |
| Novita AIfp8 | $0.99 | $2.97 | 68 tok/s | 1.98 s |
| GMI Cloudfp8 | $1.06 | $3.17 | 51 tok/s | 2.71 s |
| DekaLLM | $0.88 | $3.80 | 88 tok/s | 1.32 s |
Qwen3.8 Max (0902) providers
Sources and method
Prices are USD per 1M tokens at the standard service tier in the default region: the cheapest qualifying provider per model, with flex, batch, priority and fast endpoints and unusually low outliers never used as the headline. “Blended” weights input and output 3:1. Prices: OpenRouter, observed at the times on each model page. Speed: OpenRouter, previous 30 minutes. Usage: OpenRouter rankings, latest full UTC day (CC BY 4.0).