DeepSeek V4.1 Flash as a Labs experiment, open for a short test window: temporary, not a permanent id. DeepSeek's latest flash model, served from the official DeepSeek-V4.1-Flash open weights: a 552B mixture-of-experts (8B active per token on prefill, 16B on decode) built for fast agentic coding, with native image understanding and a 1M-token context. Reasoning has four modes: non-think (none), think low (low), think high (high, the default) and think max (max). Access is seat-gated through the Labs page while an experiment is live. It is offered at limited capacity and low availability, so expect it to be flaky and to go down under load: crash it, give it a moment, and try again. For production work we recommend the pay-per-token umans-deepseek-v4.1-flash.
Umans DeepSeek V4.1 Flash (lab) Playground Retired
playground experiment; the DeepSeek V4.1 Flash lab ran Sep 10 to Sep 19, 2026; its metrics before the Sep 18 release carry on the released model's status page as its pre-release period
retired Sep 19, 2026 · replaced by umans-deepseek-v4.1-flash · weights ↗
207.9tok/s
throughput · p50 · whole period
1.52s
TTFT · p50 · whole period
70.00%
uptime · whole period
Trends
Sep 10, 2026retired Sep 19, 2026Sep 21, 2026
Sep 10, 2026retired Sep 19, 2026Sep 21, 2026
Changelog
Sep 192026
Playground closed: Umans DeepSeek V4.1 Flash (lab) Testing
The DeepSeek V4.1 Flash lab closed on September 19, 2026. It opened September 10 and ran past the announced September 11 close. The model graduated on September 18 as the pay-per-token umans-deepseek-v4.1-flash, and the lab's metrics up to that release live on its status page as the pre-release period. Thanks to everyone who tested it.
Sep 102026
New Labs experiment: Umans DeepSeek V4.1 Flash Testing
A new lab opened on umans-deepseek-v4.1-flash-lab: DeepSeek-V4.1-Flash served from DeepSeek's official open weights, the V4.1 refresh of the V4 Flash line: a 552B mixture-of-experts (8B active per token on prefill, 16B on decode) built for fast agentic coding, with native image understanding and a 1M context window. Reasoning has four modes: non-think, think low, think high (the default) and think max. Free and seat-gated while the experiment runs, served at limited capacity and low availability: it is a lab, so expect it to be flaky and to go down under load. Crash it, give it a moment, and try again. The window closes September 11, 2026.