Mistral AI Unifies Reasoning and Coding with Mistral Medium 3.5 and Remote Agents

Apr 30, 2026 · Updated May 8, 2026

Mistral AI released Mistral Medium 3.5, a flagship 128B dense model (where all parameters activate for every request) with a 256k context window. It merges instruction-following, reasoning, and coding into one set of weights with configurable reasoning effort. The model is the new default for Le Chat and Mistral Vibe.

Model size: 128B parameters
Architecture: Dense
Context window: 256k tokens
Pricing (input): $1.50 per million tokens
Pricing (output): $7.50 per million tokens
License: Modified MIT (open weights)
Availability: Le Chat Pro, Team, Enterprise

This release signals a shift toward agentic engineering, moving beyond raw weights to managed runtimes. By enabling remote parallel agents and session teleportation—moving live coding states from local to cloud—Mistral is competing with DeepSeek's agentic coding focus and GitHub's GPT-5.5 integration. It bridges chat and execution.

Access Mistral Medium 3.5 via API at $1.50 per million input tokens or download the open weights on Hugging Face. The new Work mode in Le Chat is available for Pro, Team, and Enterprise users, enabling research tasks that pull context from connected documents, email, and calendars.

View the full update on mistral.ai

Mistral Vibe

@mistralvibeApr 29

Mistral Medium 3.5, a new flagship model in public preview by @MistralAI that merges instruction-following, reasoning, and coding into a single 128B dense model with a 256k context window and configurable reasoning effort. It's a new default model for Mistral Vibe and Le Chat. https://t.co/vcW1GaB3s1

79647

View on X

Still wondering? A few quick answers below.

Mistral Medium 3.5 is a flagship 128B dense model—where every parameter is used for every calculation—designed to handle instruction-following, reasoning, and coding. It features a 256k context window and introduces configurable reasoning effort, allowing users to adjust the model's internal deliberation. It is the new default model for Mistral Vibe and Le Chat.

The Mistral Medium 3.5 API is currently live and priced at $1.50 per million input tokens and $7.50 per million output tokens. This pricing applies to developers using the model through Mistral's API platform. The model is also available to subscribers on the Le Chat Pro, Team, and Enterprise plans.

Mistral Medium 3.5 is available as an open-weight model on Hugging Face under a modified MIT license. While the weights are publicly accessible for download and local deployment, the specific terms of the modified license govern its use. This release allows developers to run the 128B dense model on their own infrastructure or private clouds.

Mistral Vibe now supports launching remote agents in parallel to handle coding tasks. A key feature is session teleportation, which allows developers to move a local coding session to the cloud mid-way. This transfer includes the full task state, history, diffs, and approvals, enabling a seamless transition from local development to remote execution.

Work mode is a new agentic interface in Le Chat designed for complex, long-horizon tasks like research and analysis. It uses default connectors to pull context from a user's connected documents, emails, and calendars. The agent can perform multi-turn actions across these tools and maintain persistent sessions until the assigned task is complete.