MISTRAL MEDIUM 3.5
Released April 2026 (v26.04). Optimized for agentic and coding workloads. 262K context. Native vision. $1.50/$7.50 per million tokens.
Mistral Medium 3.5 (released April 2026 as v26.04) is Mistral's flagship on Council AI, deliberately tuned for agentic workflows and coding. It runs a 262K context window, accepts image input, and is priced at $1.50 per million input tokens and $7.50 per million output. Mistral retired Mistral Large in September 2026, which makes Medium 3.5 the Mistral seat in any council. The case for: agent loops that need reliable tool-calling, code-generation pipelines, long multimodal documents that fit in a single 262K pass, and keeping a European lab in the room. The case against: the hardest multi-step reasoning, where Claude Opus 5 and GPT-5.6 Sol still win, and anything that needs more than 262K of context. Web search is disabled in Council AI for Mistral models.
Medium 3.5 is the Mistral seat on Council AI. Mistral retired Mistral Large in September 2026, so Medium 3.5 inherits the flagship slot: it is the model to pick when you want a European lab in the room, strong tool-calling, and vision on a 262K window. For the hardest multi-step reasoning, put Claude Opus 5 or GPT-5.6 Sol alongside it; for code specifically, Codestral is the cheaper Mistral option.
April 2026 release; Mistral rebuilt the tuning specifically around agentic tool-use and code generation. The model card emphasizes function-calling reliability and lower hallucination on tool arguments.
Luna is the cheaper of the two by a wide margin and carries native web search through the Responses API. Medium 3.5 answers with a larger context window, vision input, and tuning aimed squarely at agentic tool use.
No. It is the active Mistral model in Council AI, and since Mistral retired Mistral Large in September 2026 it is the flagship Mistral seat. (Mistral's Magistral Medium / Small were deprecated separately.)
$1.50 per million input tokens and $7.50 per million output. Council AI bundles it inside monthly plan budgets — no separate Mistral key required.
Yes. Pair it with GPT-5.6 Sol for tool reliability and Claude Sonnet 5 for code quality, and you get three labs on three continents answering the same prompt.