Mistral Small 3.2 (24B) Released with Improved Function Calling and 128K Context
Mistral has released Mistral Small 3.2, a 24-billion-parameter open-weight model now available on Hugging Face under an Apache 2.0 license. The update significantly improves function calling reliability by reworking tool schema handling, fixing issues with optional parameter parsing that caused hallucinations in version 3.1. The model supports a 128K context window with sliding attention enabled by default in vLLM 0.6.3 and above, improving throughput for long-context workloads. Benchmark testing places it within 3–5% of GPT-4o mini on most reasoning tasks, with multi-turn tool-use accuracy rising from 71% in Small 3.1 to 78% in the new version. Quantized builds, including 4-bit GGUF variants, are already available, making the model accessible for local development on hardware such as Apple's M2 Ultra Mac Studio.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in