Deepseek's new flagship open model is built for agent loops, not benchmarks. v4.1 flash, weights…
deepseek's new flagship open model is built for agent loops, not benchmarks. v4.1 flash, weights out september 10 under mit, activates just 16b of its 552b params per token and cuts kv cache to a quarter of v4 flash while scoring 90.6 on terminal-bench 2.1.
the memory budget is the new spec sheet.
Context
DeepSeek's news post of 10 September 2026 and the Hugging Face model card describe V4.1 Flash as a 552B parameter MoE with 8B active for input and 16B for output, a KV cache at one quarter of the HBM and one eighth of the SSD versus the previous generation, and repository and weights under the MIT License. The card's table shows 90.6 on Terminal-Bench 2.1 using the Minimal mode of the DeepSeek Harness at 1M context, with other values (88.0, 84.1, 85.0, 86.1, 90.3, 85.8) across scaffolds. DeepSeek calls it the smallest model in the new architecture family and says V4.1-Pro is still to launch.
The figures are first-party and vendor-reported. 16B active per token is the output-side figure and the input side is 8B. A quarter of V4 Flash matches the HBM figure only. The 90.6 is configuration-specific. Flagship is not DeepSeek's wording, since a Pro tier is pending. The weights date of 10 September is partly unverified, because the card's own date was not stated in the text read. The memory budget is the new spec sheet is the author's take.
Watch next
- V4.1-Pro and independent Terminal-Bench 2.1 runs.
Sources
- DeepSeek V4.1 Flash (DeepSeek news, 10 Sep 2026)deepseek.com
- DeepSeek-V4.1-Flash (Hugging Face)huggingface.co
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 23:34 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →