DeepSeek V4.1-Flash is generally available today, September 10, and the weights are already yours: the release shipped with MIT open weights on day one, no API-only teaser window attached. Hosted pricing runs around $0.15/$0.60 per million tokens, with provider tables ranging $0.14-$0.15 in and $0.28-$0.60 out.

It follows V4-Pro (August 13) and lands in the middle of the densest flagship month of the year: Gemini 3.8 Flash shipped September 2, GPT-6 Astra opened to limited organizations September 3, and Grok 4.7 and Claude Opus 5.5 are due within the fortnight. Against that backdrop, V4.1-Flash is not pitched as the absolute frontier, it is the cost-per-point play, and on cost per point nothing else is close.

Why day-one MIT weights matter

MIT licensing means self-hosters can take the model as-is without negotiating terms, which is why DeepSeek releases have dominated self-hosting shortlists all year. DeepSeek's API lists a 1M-token context for the current Flash model (an earlier version of this story said 128K, which was wrong), so the context gap with Western flagships is not the trade-off it first looked like; cost-sensitive volume workloads are the use case. DeepSeek's API has since moved to the V4.1-Flash-1017 checkpoint, at $0.15/$0.60 off-peak and twice that at peak hours.

At this price, and with weights in hand, there is nothing to wait for, our Model Guide gives DeepSeek an Adopt Now badge.

For European self-hosters, the next data point is Mistral Large 4, which entered preview October 6 with weights promised by month-end. Until then, V4.1-Flash is the open-weight release every other September launch gets priced against.