mistral.ai
|
ksl
|
|
Mistral released Small 4, a 119B-parameter mixture-of-experts model with only 8B active per token, under Apache 2.0. It merges what used to be three separate Mistral products – Magistral for reasoning, Pixtral for vision, and Devstral for coding – into a single model with configurable reasoning effort. The 256k context window and 3x throughput improvement over Small 3 make it a direct play for the high-volume API tier. On LiveCodeBench it outperforms GPT-OSS 120B while generating 20% less output, which matters when you’re paying per token. OpenAI just shipped GPT-5.4 nano this same week, and Mistral is clearly angling for the same developer budget – open weights and Apache licensing remain their strongest differentiator in that fight.
