DeepSeek updates V4.1-Flash with new coding scores and Search tools
DeepSeek’s October 17 refresh offers downloadable MIT-licensed weights, a one-million-token context window and reported gains on coding benchmarks; the scores are company-run and setup-dependent.
2 min read
DeepSeek updated its V4.1-Flash model to V4.1-Flash-1017 on October 17, describing it as a minor revision aimed at reasoning, coding, agent tasks and visual benchmarks. The model card lists 763 billion total parameters, a one-million-token context window, multimodal image-text support and weights under the MIT License. The weights are downloadable, while DeepSeek’s hosted API continues to serve the latest version through the existing deepseek-flash model name.
DeepSeek’s comparison table reports a Terminal-Bench 4.0 score of 42.3, up from 31.2 for the previous Flash version, and DeepSWE v1.1 at 77.0 versus 74.2. These are vendor-reported results, not an independent replication. DeepSeek says it used the Claude Code harness for Terminal-Bench and Mini-SWE-V2 for DeepSWE, so the scores should be read in the context of those setups rather than as a universal ranking.
The release also includes server-side Search and Fetch tools for API users, with a free period through October 31 for use via DeepSeek Harness, Claude Code and Codex. The linked deepseek-recipe project supplies Rust libraries and Python bindings for encoding prompts and parsing responses; GitHub showed 411 total stars and 38 forks on October 18. That is a point-in-time count, not evidence of recent growth.
The model card offers local inference instructions, but this is a 763B-parameter download and Hugging Face did not list a hosted inference provider when checked. Practical deployment therefore depends on suitable hardware or API access.
Photo by Paul Hanaoka on Unsplash, used under the Unsplash License.
Sources: DeepSeek release notes · DeepSeek-V4.1-Flash-1017 model card and benchmark table · deepseek-recipe on GitHub · Photo by Paul Hanaoka · Unsplash License
Conversation
0 approved commentsNo comments yet. Start a thoughtful conversation.
Join the conversation.
Sign in with ChatGPT to leave a comment or rate this article.