DeepSeek published DeepInfer on GitHub under Apache 2.0 on Oct. 20 and said it is the offline inference stack behind its production services. The README supports only DeepSeek-V4.1-Flash, on Nvidia SM100 GPUs and Huawei Ascend 950 NPUs, and it leaves out a KV-cache system, prefill/decode disaggregation, and replica load balancing. The speed charts are the company's measurements on 4 SM100 GPUs for prefill and 64 for decode. The Next Web reported that Bloomberg had not found an independent test of those gains.
/ Sources
/ About this story
Compiled by Venture Atlas from the sources above, using automated AI-assisted research. This is a summary of reporting published elsewhere, not original reporting - follow the source links for the full account. See our editorial standards.
Something wrong here? Email apostructura@gmail.com and we will fix it.
/ Related
- DHS orders civilian agencies to remove DeepSeekMonday, October 19, 2026
- DeepSeek upgrades V4.1-Flash and keeps the same API priceSaturday, October 17, 2026
- DeepSeek opens Search and Fetch APIs at $0.75 per 1,000Saturday, October 17, 2026
- DeepSeek open-sources Ascend tools with Huawei supportWednesday, September 30, 2026