DeepSeek Opens Its Engine, Keeps the Plumbing
DeepSeek open sourced the engine behind V4.1-Flash and its 8 GPU-hour revenue, but kept the cache and load-balancing code a full service needs.
Read the storyProduct · General assistants
DeepSeek’s conversational application for questions, reasoning and coding assistance.
Get AI Secret’s free daily briefing, including DeepSeek Chat and the wider AI industry.

DeepSeek’s conversational application for questions, reasoning and coding assistance.
Official product information checked .
Briefs, articles and relevant sections, newest first. Explore older coverage below.
More coverage
DeepSeek open sourced the engine behind V4.1-Flash and its 8 GPU-hour revenue, but kept the cache and load-balancing code a full service needs.
Read the storyMiniMax revealed the anonymous Space Bunny model as M3.1 Flash, a cheaper update it says cuts agent task costs 70%, with self-reported benchmark scores.
DHS barred DeepSeek from civilian agencies under a July order whose ban on anything "developed using" DeepSeek no vendor can verify, a security expert says.
Plus: Anthropic Pulled Up the Ladder, OpenAI Copied Meta's Ad Machine
DeepSeek's V4.1 Flash tops open models on Code Arena WebDev at a tenth of Kimi K3's cost, though Claude Opus 5 still leads most benchmarks.
Plus: The Dustbin Is Full of Mathematicians, Anthropic Priced the Demo
Plus: Mistral Knows What It Is, First AI Film Festival Crowns Its Winner
Plus: OpenAI and Anthropic Are Cutting Air, The Storefront Moved Into the Chat
Three labs asked the industry to slow down. The market has already answered.
Plus: Your Tumor Picks the Cure, The Robots Have Never Had a Job