Mistral (Medium 3.5)

What it is

Mistral is the French AI lab's model lineup, aimed at the same general-purpose assistant use cases as GPT, Claude, and Gemini. Mistral is somewhat unusual among major labs in that its flagship-tier models are closed weight while several of its smaller and mid-sized models are released openly — giving developers a path from a fully open, self-hostable model up to a more capable hosted flagship from the same lab.

Current lineup

Mistral ships across several tiers rather than a single model: Mistral Medium 3.5 (April 2026) and Mistral Small 4 (March 2026) are the current general-purpose releases, alongside Voxtral TTS and Voxtral Realtime, dedicated voice models for speech synthesis and real-time voice applications.

Where these numbers come from

The HumanEval figure above is from Mistral Large 2 (24.11), the lab's prior flagship generation — this entry is a seed example and should be updated via the admin dashboard with official benchmarks for the current Medium/Small generation as they're published.

Notable traits

• Strong particular emphasis on coding and multilingual performance relative to model size.
• European-headquartered (Paris), often discussed in the context of EU AI regulation and digital sovereignty conversations, distinct from the US-headquartered labs that dominate this list.
• Available directly via Mistral's API as well as through major cloud marketplaces.