Pyyan / Compare / Mistral Large 4 vs DeepSeek V4.1 Flash

Mistral Large 4 vs DeepSeek V4.1 Flash

2 of 5

Open-Weight Models · verified 11 Nov 2026

Mistral Large 4Mistralcurrent
DeepSeek V4.1 FlashDeepSeekcurrent
3 slots left
SpecificationMistral Large 4DeepSeek V4.1 Flash
SummaryA trillion parameters with the weights out, and a licence that charges the hosts.Twice the parameters of the Flash it replaces, and cheaper to run.
Context1M1M
Input ($/Mtok)$1.36$0.3
Output ($/Mtok)Not published$1.2
LicenceMistral Ecosystem Open-Weight License (modified MIT)MIT
Max outputNot published256K
Parameters1T total, 49B active (52B with embeddings)552B total, 8B / 16B active
Released6 Oct 202610 Sep 2026, refreshed 17 Oct
Model IDmistral-large-4deepseek-flash
CategoryOpen-Weight ModelsOpen-Weight Models
OfficialMistral ↗DeepSeek ↗

Highlighted rows are where these differ.

Mistral Large 4

  • Weights published on 29 October 2026 under the Mistral Ecosystem Open-Weight License, a modified MIT: a business offering the model as a service needs a separate licence from Mistral
  • Mistral reports 48% on the Artificial Analysis Coding Agent Index, tied with GLM-5.3 at 47% fewer tokens, and 41% on Terminal-Bench 4.0, its own figures
  • Sources disagree on size: the Hugging Face card says 1T total and 49B active (52B with embeddings), other coverage 1.05T and 52B
  • 1M token context, with a 1.6B parameter vision encoder

Best for self-hosting a European frontier model.

Full spec sheet →

DeepSeek V4.1 Flash

  • 552B total, 8B active while reading the prompt and 16B while generating, with a 1M token context
  • DeepSeek says it beats V4-Pro; V4-Pro requests route to it from 14 September 2026
  • MIT licence, native image input, and a KV cache about a quarter the size of V4-Flash's
  • Refreshed on 17 October as V4.1-Flash-1017, still MIT, improving on all eleven benchmarks DeepSeek published

Best for cheap agentic coding with images.

Full spec sheet →