Mistral AI Launches Mistral Medium 3.5, Consolidating Coding and Reasoning Models
A single 128B open-weight model now handles reasoning, coding, and instruction-following.
Mistral AI unveiled Mistral Medium 3.5 on April 30, 2026, a 128-billion-parameter open-weight model that consolidates instruction-following, reasoning, and coding into a single system. This flagship model replaces both the dedicated reasoning model Magistral and the coding model Devstral 2, streamlining Mistral's lineup for developers and enterprises. By integrating these capabilities, Mistral Medium 3.5 aims to handle complex multi-turn tasks without switching between specialized models. The open-weight release allows organizations to fine-tune and deploy the model on their own infrastructure, offering flexibility for proprietary use cases. Early benchmarks suggest competitive performance against proprietary models in code generation and logical reasoning while maintaining strong instruction adherence.
Beyond the model, Mistral updated its developer tool Vibe with asynchronous cloud agents, enabling long-running background tasks like automated testing or data processing. The Le Chat assistant gains a new 'Work mode' that supports multi-step workflows, allowing users to chain commands and reuse context across sessions. These ecosystem enhancements position Mistral Medium 3.5 not just as a model but as part of a broader platform for AI-assisted development and productivity. Together, the release simplifies the toolchain for developers while giving end-users more powerful, context-aware assistance.
- Unifies instruction-following, reasoning, and coding in a single 128-billion-parameter open-weight model
- Replaces two former models: Magistral (reasoning) and Devstral 2 (coding)
- Vibe tool adds async cloud agents; Le Chat introduces Work mode for multi-step workflows
Why It Matters
Consolidates two core capabilities into one model, simplifying enterprise deployment and reducing infrastructure overhead.