Skip to content
EVENT / TOPICHistorical

DeepSeek Open-Sources DeepInfer Inference Engine, Claims 155T Tokens/Day Processing

1 reports1 sources10/21/2026 updated

Current overview (not historical)

AI overview

DeepSeek has open-sourced the DeepInfer inference engine at PyTorch Conf, claiming a single cluster can process 155T tokens daily, with an estimated revenue of $38 per card per hour. The engine supports platforms including NVIDIA SM100 and Huawei Ascend 950, offering features like distributed inference, speculative decoding, and SWA Bounded Replay. DeepSeek stated that future model inference implementations and optimizations will also be shared in the repository. (Note: One report mentions a $22m/day revenue projection at an ~80% margin, leading to an 81-day payback period for a specific setup.)

Written with AI from the reporting · updated 10/23/2026

Latest developmentOct 21, 14:54
DeepSeek open-sources DeepInfer inference engine, claims 155T tokens/day at 80% margin
Explore the timeline, relations and evidence
Expand the detailed fact route and sources

Story route

A station-by-station route from the first public evidence onward; every station carries its own evidence and uncertainty.

  1. 1Unclassified fact stationOct 21, 00:00recorded Oct 21, 14:56single sourcerevised

    DeepSeek DeepInfer开源，单集群155T tokens/day，$38/accelerator-hour

    Read original source ↗ Latent SpaceReplay this moment in the explorer
    Evidence & revisions · 1 source records
    Subject：
    deepseek
    Action：
    open source DeepInfer inference engine
    Object：
    DeepSeek V4.1 Flash inference framework
    Occurred (fact side)：
    Oct 21, 00:00
    Latent Space ↗Report pageRevision 4 · currently revised
    Published (valid side)：
    Oct 21, 14:54
    First seen (known side)：
    Oct 21, 14:56
    Fact stationSingle source · one-sided claimRevised

    Copy may have been revised later; time replay is bounded. Layer labels group reading only, not causation between layers.

Uncertainty marks flag where to check the sources; they are not proven causation.

Report timeline

Explore every side through dated source reports.

Oct 21
  1. Latent SpaceSelectedRelated fact: single sourceSource report revised
    DeepSeek open-sources DeepInfer inference engine, claims 155T tokens/day at 80% margin

    Report first recorded Oct 21, 14:56

    DeepSeek released DeepInfer, the inference engine behind V4.1 Flash, as open source on the first day of PyTorch Conference. The engine supports distributed inference, speculative decoding, and structured output on NVIDIA and Huawei Ascend platforms. According to DeepSeek's translated blog post, their deployment achieves 155 million tokens per day with a claimed inference margin exceeding 80%, generating approximately $22 million in daily revenue at off-peak prices. The architecture combines vLLM with SGLang and TRRTLLM, though DeepSeek notes their most ultra-optimized KV cache and load balancing components remain proprietary.

    Read original source ↗

Event discussion

Reader opinions are hosted on GitHub Discussions; posting requires a GitHub account. Comments are not verified facts.

Open GitHub Discussions