Independent newsroom

Tuesday, October 20, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

AI · Tuesday, October 20, 2026

DeepSeek releases the software it uses to run its models

The File desk · Oct. 20, 2026, 6:44 a.m. ET

Partially confirmed2 primary sources

Status

Partially confirmed. The GitHub repository was re-opened for this edition and carries an Apache-2.0 license. The announcement quote is the company's post, read by the morning desk and not re-fetched here.
  • Confirmed
    A public repository named DeepInfer describes itself as DeepSeek's inference framework and is licensed Apache-2.0.
  • Partially confirmed
    The 5:30 a.m. ET announcement, including the claim that it powers DeepSeek-V4.1-Flash, is the company's post.
DeepSeek said Tuesday morning that it has published the software it uses to run its models. The repository is called DeepInfer. The license on the page this desk opened is Apache-2.0.1
The GitHub page says, “DeepInfer is the high-performance inference framework that powers DeepSeek's large-scale production services.” It says the release “provides offline inference with the core features needed for serving at scale.” The morning desk's reading of the company's post, at 5:30 a.m. ET, says the software is “Powering DeepSeek-V4.1-Flash at scale.” That sentence is the company's.1
The same desk read the repository's creation date as Oct. 16, with the announcement today. A second repository, for token encoding, wasn't opened.
Why it matters (Estimate). Running a model, after it is trained, is a large share of the cost of offering it to users. Publishing the production software can make it cheaper for other people to run an open model. Whether this release does that is a claim, not a measurement this desk made.1
The announcement post wasn't re-fetched on this pass. The repository page was.
How we know

Notes

  1. 1.GitHub, deepseek-ai/DeepInfer, re-opened for this edition. https://github.com/deepseek-ai/DeepInfer “DeepInfer is the high-performance inference framework that powers DeepSeek's large-scale production services. This open-source release provides offline inference with the core features needed for serving at scale.” License Apache-2.0. The AI desk read the GitHub API as creating the repo on Oct. 16 and the announcement as today. Company post, 5:30 a.m. ET Tuesday, not re-fetched on this pass: https://x.com/deepseek_ai/status/2112476496283775135 “Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency.” A second repo, https://github.com/deepseek-ai/deepseek-recipe wasn't opened.

What is still unknown or disputed

  • The announcement post wasn't re-fetched.
  • The companion recipe repository wasn't opened.
  • No independent speed test was found.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    DeepInfer repository

    DeepSeek · October 20, 2026

    DeepInfer is the high-performance inference framework that powers DeepSeek's large-scale production services.

    https://github.com/deepseek-ai/DeepInfer

  2. Source 2 · fetch incomplete

    DeepSeek announcement of DeepInfer

    DeepSeek · October 20, 2026

    Introducing DeepInfer, our in-house inference framework, now fully open source.

    https://x.com/deepseek_ai/status/2112476496283775135

Ask this story

Answers come only from this story and its sources. Free test: limited questions per day.

0/300