← October 18, 2026 briefing · added in the 12:00 KST update

P2 Model Open weights Language modelsAgentsFollow-up12:00 updateDeepSeekTencent

DeepSeek V4.1-Flash-1017 follow-up: official release confirmed with MIT-licensed weights, rolled out to web, app and API at unchanged prices (report)

Announced October 17, 2026

Verified (fact)

'DeepSeek-V4.1-Flash-1017', which had earlier surfaced through the community, was officially released on the 17th. It is a minor update improving on V4.1 Flash, released last month, and has been applied to the website, app and API, with weights released under the MIT license. API pricing is unchanged. Pandaily reported that task execution time and output length were reduced and that the model was tuned for DeepSeek Harness and Tencent's CodeBuddy and WorkBuddy.

Needs fact-check (claim)

DeepSeek said benchmark scores rose and the tokens needed to complete tasks fell compared with the previous version. It claims that in its own evaluations the model outperformed OpenAI and Anthropic models on some software development and automation tasks.

Why it matters (interpretation)

The approach lowers real-world usage costs by cutting the number of tokens needed while keeping unit prices the same. MIT-licensed weights that can be used freely for commercial purposes also matter in the coding agent market.

Confidence medium · official source pending

Sources

Story thread

  1. 2026-10-17 DeepSeek releases 'V4.1-Flash-1017' on Hugging Face; Terminal-Bench 4.0 rises from 31.2 to 42.3 (community reports)
  2. 2026-10-17 DeepSeek V4.1 Flash 1017 follow-up: company's first CED (Causal Encoder-Decoder) architecture, native image understanding; 552B total, 8B input/16B output active (community reports)
  3. 2026-10-18 DeepSeek V4.1-Flash-1017 follow-up: official release confirmed with MIT-licensed weights, rolled out to web, app and API at unchanged prices (report)
  4. 2026-10-19 DeepSeek V4.1-Flash-1017 follow-up: independent Terminal-Bench 4.0 scores of 34% and 31.8% trail the model card's 42.3% (community reports)

More about DeepSeek