Mistral AI has announced Mistral Medium 3.5—the first flagship model to combine instruction following, reasoning, and coding capabilities into a single 128-billion-parameter dense neural network. This update does more than just improve performance; it transforms the approach to working with AI.
What's new?
- Autonomous cloud coding: Agents now operate in parallel and asynchronously—you launch a task from the CLI or Le Chat, and the AI executes it independently, notifying you upon completion.
- Work Mode in Le Chat: A new mode for complex tasks—such as research, analytics, and cross-tool workflows. The model selects the tools itself, handles errors, and sees the task through to completion.
- 256K context: Handling long dialogues, complex codebases, and multitasking without loss of quality.
- Open weights: The model is available under the Modified MIT License — it can be deployed on just 4 GPUs.
- Pricing: $1.5 per 1M input tokens, $7.5 per 1M output tokens (on Pro, Team, and Enterprise plans).
For whom?
Developers: Automation of routine tasks (refactoring, tests, dependency upgrades).
Commands: Parallel execution of tasks without a dependency on the local terminal.
Enterprises: Integration with GitHub, Jira, Slack, and Linear — the AI works as a full-fledged team member.
Where can I try it?
The model is already available in Mistral Vibe, Le Chat, and via the API. Open weights are available on Hugging Face.
Source: mistral.ai
0 Comments