DeepSeek Launches DeepSeek-V4 Preview as Open Source with 1M Context

ArticleDeepSeek AIAnnouncement

DeepSeek has open-sourced the preview release of DeepSeek-V4. The line-up features DeepSeek-V4-Pro and DeepSeek-V4-Flash, standardising a 1M context window and introducing architectural innovations to cut compute and memory overhead.

Key points

  1. DeepSeek-V4-Pro features 1.6T total and 49B active parameters, delivering competitive performance in agentic coding, math, and STEM benchmarks.
  2. DeepSeek-V4-Flash operates with 284B total and 13B active parameters, offering lower latency and cost-effective API access.
  3. A novel attention mechanism combining token-wise compression and DeepSeek Sparse Attention (DSA) powers a standard 1M-token context length.
  4. Optimised for agentic workflows, the release integrates with tools like Claude Code, OpenClaw, and OpenCode.
  5. APIs support both OpenAI ChatCompletions and Anthropic schemas alongside Thinking and Non-Thinking modes; older models will be retired on 24 July 2026.

Why it matters

By providing open weights alongside 1M-token context windows at reduced operating costs, DeepSeek presents a direct alternative to leading proprietary models for agentic software engineering and reasoning tasks.

Analysis

The architectural reliance on DeepSeek Sparse Attention highlights ongoing efforts to make ultra-long contexts computationally viable in production. By supporting standard API formats from OpenAI and Anthropic, DeepSeek simplifies migration paths for developers while focusing heavily on automated coding and agent tooling.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • X post:Mia

    DeepSeek Releases Official DeepSeek-V4-Flash-0731 Weights on Hugging Face

    DeepSeek has officially published the weights for DeepSeek-V4-Flash-0731 under an MIT licence on Hugging Face. The release supersedes the preview version, features enhanced agentic capabilities, and incorporates an attached speculative decoding module.

    336bookmarks153.4kviews

    AI & LLMs· Announcement

  • Article:DeepSeek

    DeepSeek API Changelog: Launch of V4-Flash-Vision-Exp and V4-Pro GA

    DeepSeek's API changelog details the introduction of the multimodal DeepSeek-V4-Flash-Vision-Exp model alongside the GA release of DeepSeek-V4-Pro. The V4 family introduces native support for the Responses API format, granular thinking effort levels, and off-peak pricing.

    AI & LLMs· Announcement

  • Link:Tencent Hy Team

    Tencent Releases Hy4-preview MoE Model Under Apache 2.0 License

    The Tencent Hy Team has released Hy4-preview, an open-source Mixture-of-Experts (MoE) model featuring 770B total parameters and 49B activated parameters per token. The flagship model supports a 1M token context window and is tuned for software engineering, office analytics, and scientific research.

    AI & LLMs· Announcement

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.