DeepSeek open-sources DeepInfer, its production inference engine
Apache 2.0, one model, two kinds of hardware, and no KV cache or load balancing.
On October 20, 2026, DeepSeek published DeepInfer on GitHub under the Apache 2.0 license. The README calls it "the high-performance inference framework that powers DeepSeek's large-scale production services." This release covers offline inference for one model.
What changed
- Repository:
deepseek-ai/DeepInfer, Apache 2.0. - Supported model: DeepSeek-V4.1-Flash only. The default checkpoint is
deepseek-ai/DeepSeek-V4.1-Flash-1017. - Supported hardware: NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs.
- Listed features include optimized kernels (FlashMLA, DeepGEMM, DeepEP), attention DP and MoE expert parallelism, draft-token decoding, batch-invariant inference and deterministic sampling (NVIDIA only), and structured outputs (regex, JSON schema, EBNF).
- Not included: a KV cache system, prefill and decode disaggregation, and load balancing across replicas. DeepSeek lists all three as required for a complete inference service.
What it means
Running V4.1 Flash on DeepSeek's own stack needs SM100 or Ascend 950 hardware, Linux, Python 3.12, and CUDA 13 or CANN 9.2, plus your own caching and load-balancing layer. For most teams, a hosted endpoint is still the shorter path.
Hosted prices for DeepSeek V4.1 Flash 1017
RouterPlus catalog, November 1, 2026, per million tokens.
| Provider | Input | Output |
|---|---|---|
| Sail Research | $0.09 | $0.30 |
| Wafer | $0.13 | $0.50 |
| RunInfra | $0.15 | $0.60 |
| Doubleword | $0.15 | $0.60 |
| Particle AI | $0.2125 | $0.85 |
DeepSeek's own API lists deepseek-flash (now V4.1-Flash-1017) at $0.30 input and $1.20 output at peak, and half that off-peak (DeepSeek pricing page, November 1, 2026).
See DeepSeek V4.1 Flash 1017 and DeepSeek V4.1 Flash for every provider.
What to do
To call the hosted model through RouterPlus with one key:
curl https://api.routerplus.com/v1/chat/completions \
-H "Authorization: Bearer $ROUTERPLUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "deepseek/deepseek-v4.1-flash-1017", "messages": [{"role": "user", "content": "Hello"}]}'Sign up at app.routerplus.com/login for an API key. Marketplace calls use prepaid credits at the listed rates.