2 versions from Mistral AI, reviewed by the TopReviewed AI panel.
| Model | Score | Status | Released | Price in/out | Context | |
|---|---|---|---|---|---|---|
| Mistral Medium 3.5 open-weight agentic coding at sub-frontier price | 8.4 | GA | 2026-04-28 | $1.50 / $7.50 | 256K | Review → |
| Mistral Medium 3.1 cheap multilingual multimodal chat at volume | 7.7 | GA | 2025-08-12 | $0.40 / $2 | 131K | Review → |
Mistral Medium 3.5 (released 29 April 2026) is Mistral's most strategically important model of the year: a single 128B dense set of weights that merges what used to be three separate models — chat (Medium 3.1), reasoning (Magistral), and the agentic coder (Devstral 2) — into one SKU with a configurable reasoning effort knob. It scores 77.6% on SWE-bench Verified, within two points of Claude Sonnet 4.5, at $1.50/$7.50 per 1M tokens, and ships under a modified-MIT license that is open for nearly everyone except large-revenue enterprises. The buyer's sentence: the best open-weight agentic coding model at the Medium tier, if you can live with a license carve-out and a high output price. - Provider: Mistral AI (Paris, France) - Release: 2026-04-29, status GA - Context: 256,000 tokens; max output 32,768 - Modalities: text + image in, text out (native multimodal) - Knowledge cutoff: ~February 2026 - Headline price: $1.50 input / $7.50 output per 1M tokens - Architecture: dense 128B with `reasoning_effort` toggle (none/high)
Full review →Mistral Medium 3.1 (release 25.08, shipped 13 August 2025) is the cost-optimised workhorse of Mistral's proprietary "Premier" tier: a closed-weight, multimodal, 128K-context model priced at $0.40/$2.00 per 1M tokens — among the cheapest frontier-adjacent multimodal models on the market. It is an incremental refinement of Medium 3 (May 2025) with better instruction following, sharper tool use, and improved vision. The buyer's sentence: when budget is the dominant constraint and the workload is chat/Q&A rather than agentic coding, this is still the value pick — though Medium 3.5 now supersedes it for coding and agentic work. - Provider: Mistral AI (Paris, France) - Release: 2025-08-13, status GA - Context: 131,072 tokens; max output 16,384 - Modalities: text + image in, text out (native multimodal) - Knowledge cutoff: ~April 2025 - Headline price: $0.40 input / $2.00 output per 1M tokens - Architecture: dense (parameter count undisclosed — Premier/closed)
Full review →