DeepSeek V4.1-Flash Undercuts Its Own Flagship, and Takes Over Its API Name
Image: api-docs.deepseek.com
DeepSeek released DeepSeek-V4.1-Flash on 10 September 2026, with open weights on Hugging Face under the MIT licence.
DeepSeek's own models and pricing page gives the API name as deepseek-flash, with DeepSeek-V4.1-Flash as the model version. It lists a 1M context length, a maximum output of 384K tokens, and support for both thinking and non-thinking modes, with thinking the default. Vision input is supported — DeepSeek V4-Pro is listed as not supporting it.
The pricing gap between the two is the substance of the release. Per million tokens:
| V4.1-Flash | V4-Pro | |
|---|---|---|
| Input, cache hit (off-peak / peak) | $0.003 / $0.006 | $0.022 / $0.044 |
| Input, cache miss (off-peak / peak) | $0.15 / $0.30 | $0.66 / $1.32 |
| Output (off-peak / peak) | $0.60 / $1.20 | $1.98 / $3.96 |
The concurrency limit differs as well: 2,500 for deepseek-flash against 500 for deepseek-v4-pro.
The legacy API names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but DeepSeek's documentation states the models behind them have been retired and their requests are routed onward — so an application pinned to V4-Flash is already being served by the new model.
Secondary coverage reports the model as a 552-billion-parameter mixture of experts with 8B parameters active on input and 16B on output, and reports that DeepSeek claims it outperforms the larger V4-Pro. Those figures are not on the pricing page cited above and are not confirmed here.
Enjoyed this? Get more in your inbox.
Up to ten of the week's AI news stories, by email once a week.