DeepSeek releases open-weight V4.1 Flash with lower inference costs
DeepSeek released DeepSeek-V4.1-Flash, a 552-billion-parameter mixture-of-experts model with 8 billion active parameters for input and 16 billion for output. The release adds native vision, a new causal encoder-decoder design, lower cache requirements, API access, and MIT-licensed weights.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
Open weights, API availability, native vision, and reduced inference requirements broaden access to a capable agent-oriented model and increase competitive diffusion. Performance is based mainly on provider benchmarks, and the release does not itself demonstrate loss of control or a real-world incident.
Assessment history
-
R1
Toward 57 · confidence 87
Adds DeepSeek's dated V4.1 Flash release with open weights, API deployment, native vision, and lower inference requirements.
11 Sept 2026
Share this page
-
DoomBench assesses “DeepSeek releases open-weight V4.1 Flash with lower inference costs” as evidence moving toward doom, with magnitude 57 and confidence 87 out of 100 in the competitive race category.
-
The DoomBench assessment of “DeepSeek releases open-weight V4.1 Flash with lower inference costs” is based on reporting from DeepSeek and records the editorial rationale, source quality, attribution, and revision history.
-
DoomBench summarizes “DeepSeek releases open-weight V4.1 Flash with lower inference costs” as follows: DeepSeek released DeepSeek-V4.1-Flash, a 552-billion-parameter mixture-of-experts model with 8 billion active parameters for input...
https://www.doombench.com/news/deepseek-releases-open-weight-v4-1-flash-with-lower-inference-costs-2026-09-10