DeepSeek released DeepSeek-V4.1-Flash on 10 September 2026 — and cut its API price to $0.15 per million input tokens off-peak, with output at $0.60. The model has a 552 billion parameter mixture-of-experts backbone but activates only 8 billion parameters for each input token, and the weights are published on Hugging Face under the MIT license, with a one-million-token context window.

The cost story is about the KV cache — the working memory a model keeps for every token in your prompt, which makes long agent runs expensive because it grows with the conversation and has to stay in GPU memory. DeepSeek says V4.1-Flash holds 890 bytes of KV cache per token: about a quarter of what DeepSeek-V4-Flash needed, and roughly one 437th of the original DeepSeek-V1.

Two design choices do that work. A causal encoder-decoder layout splits the model's 40 layers into a 20-layer encoder and a 20-layer decoder, so the decoder builds its cache once from the encoder's output instead of maintaining one cache per layer — which is why the model activates 8 billion parameters while reading a prompt and 16 billion while writing. On top of that, DeepSeek stores the cache in FP4, a four-bit number format, and shares attention indices between layers. A reasoning-effort dial from 1 to 100 lets you trade accuracy against cost per request instead of switching models.

On benchmarks, DeepSeek's own numbers show the model matching frontier models on agent and coding tests: 74.2 on DeepSWE v1.1, 90.6 on Terminal-Bench 2.1 and 90.9 on GPQA Diamond in its published comparison table. It lags on the hardest reasoning tests — 36.8 on Humanity's Last Exam versus 56.3 for Claude Opus-5.0. All benchmark figures are DeepSeek-reported, not independently audited, and DeepSeek itself disclosed that the headline DeepSWE score is the best of eight test scaffolds.

The release is already displacing DeepSeek's own flagship. The older V4-Flash and V4-Flash-Vision-Exp identifiers are retired and now route to V4.1-Flash. And from 12:00 Beijing time on 14 September 2026, calls to deepseek-v4-pro also route to V4.1-Flash, billed at Flash prices — roughly a quarter of what V4 Pro cost. DeepSeek says a V4.1 Pro will follow but gives no date.

Sources