Krux
March 25, 2026
Mistral Merges Three AI Models Into One Open-Source Release
Published: March 25, 2026 at 12:34 AM
Updated: March 25, 2026 at 12:34 AM
100-word summary
Mistral just collapsed its reasoning, vision, and coding models into a single open-source release. Small 4 handles text, images, and code without switching between specialized tools, running 40% faster than its predecessor in low-latency mode. The 119-billion-parameter model lets you toggle between quick answers and deep reasoning depending on the task. Need a chatbot to parse invoices and answer questions? Same model. Building coding agents? Still the same model. The catch: you'll need serious hardware. Minimum entry is four NVIDIA H100 GPUs. But if you've got the compute, you can fine-tune and host it yourself under Apache 2.0. No vendor lock-in, no API fees for every query.
What happened
Mistral just collapsed its reasoning, vision, and coding models into a single open-source release. Small 4 handles text, images, and code without switching between specialized tools, running 40% faster than its predecessor in low-latency mode. The 119-billion-parameter model lets you toggle between quick answers and deep reasoning depending on the task. Need a chatbot to parse invoices and answer questions? Same model. Building coding agents? Still the same model.
Why it matters
The catch: you'll need serious hardware. Minimum entry is four NVIDIA H100 GPUs. But if you've got the compute, you can fine-tune and host it yourself under Apache 2.0. No vendor lock-in, no API fees for every query.