Latest: DeepSeek V4.1-Flash-1017 Launches Open-Weight, Beats Opus 5 on Coding Benchmark
An open-weight model with an MIT license that scores above closed competitors on a coding benchmark at a fraction of the price gives enterprises and developers a low-cost alternative for software engineering tasks. The permissive license stands out as other models move toward more restrictive terms.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
DeepSeek released V4.1-Flash-1017 on October 19, 2026, a 552B-A16B mixture-of-experts model with image input support and an MIT license. On the DeepSWE v1.1 benchmark, it scored 77.0 against 74.0 for Claude Opus 5 and 73.0 for GPT-5.6 Sol, though DeepSeek ran its own tests. The model is priced at $0.15 per million input tokens and $0.60 per million output tokens off-peak, compared to roughly $1.86 for GPT-5.6 Sol and $3.15 for Claude Fable 5 on a standard test suite. DeepSeek's Search and Fetch API tools cost $0.75 per 1,000 requests, versus Brave, Tavily, and Perplexity at $5 per 1,000 and up. Free usage through DeepSeek Harness, Claude Code, or Codex ends November 1.
A follow-up to the October 19 launch of V4.1-Flash-1017 identified the model's architecture as a 552B-A16B mixture-of-experts with image input support, and Interconnects.ai described the MIT license as notable at a time when "licenses seem to become more bespoke and exclusionary".
Databricks also said the model lands on the Pareto frontier for enterprise document reasoning among open-source options on OfficeQA Pro V2.
DeepSeek shipped V4.1-Flash-1017 on October 19, 2026, releasing the weights under an MIT license and scoring 77.0 on the DeepSWE v1.1 coding benchmark against 74.0 for Claude Opus 5 and 73.0 for GPT-5.6 Sol, though DeepSeek ran its own tests.
API pricing is $0.15 per million input tokens off-peak, versus roughly $1.86 for GPT-5.6 Sol and $3.15 for Claude Fable 5, and Search and Fetch tools are $0.75 per 1,000 requests against $5 and up from Brave, Tavily, and Perplexity. Free access through DeepSeek Harness, Claude Code, or Codex ends November 1. Databricks listed it as an endpoint on launch day, routing all calls through Unity Gateway.
BackgroundFrontier models, open models, and how they are compared
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.