Developer FAQ
Last updated 2026-09-25. The first sentence of each answer stands alone.
DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with 552B backbone parameters and a context of up to one million tokens.
The model card describes a Causal Encoder-Decoder that activates 8B parameters per token during prefill and 16B during decode, Engram conditional memory of 196B parameters, 384 routed experts with 6 activated, and a global KV cache of 890 bytes per token. On the API it is called with model deepseek-flash.
V4.1-Flash has a 552B backbone (8B active in prefill, 16B in decode) while V4-Pro has 1.6T parameters with 49B activated; both API models list a 1M context.
At peak, deepseek-flash costs $0.30 per 1M cache-miss input tokens, $0.006 per 1M cache-hit input tokens, and $1.20 per 1M output tokens; off-peak is half of each rate.
This site does not publish Claude or GPT prices, because those figures were not checked against the vendors’ current price pages.
The current model names are deepseek-flash for DeepSeek-V4.1-Flash and deepseek-v4-pro for DeepSeek-V4-Pro-0813.
Point the OpenAI SDK at base URL https://api.deepseek.com and set the model to deepseek-flash or deepseek-v4-pro.
Both deepseek-flash and deepseek-v4-pro list a 1M token context length and a maximum output of 384K tokens.
Off-peak rates are half of peak rates, and peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays.
DeepSeek does not publish a ranking for Cline, Aider, or Cursor; the current general API model is deepseek-flash.
V4.1-Flash publishes weights and an inference folder for local runs, but that README introduction does not state a minimum GPU count.
They are open-source libraries from DeepSeek: FlashMLA for MLA attention kernels, DeepEP for expert-parallel communication, and DeepGEMM for GPU GEMM kernels.
DeepSeek-R1 is the January 2025 reasoning model; the current API pricing table serves V4.1-Flash and V4-Pro, not a separate R1 model name.