dev-newsMistral AI logomistral.ai

Latest stable Mistral Python SDK: 3.2.0, released (release notes). 86 stable releases since 7 Aug 2024.

Mistral AI

52mistralai releases in 2026, two of them majors

Jan1.10.11.11.11.12.0Feb1.12.11.12.21.12.31.12.4Mar2.0.02.02.0.12.0.22.0.32.0.42.0.52.1.02.1.12.1.2Small 4LeanstralForgeSmall 4 路 Leanstral 路 Forge2.1.32.2.02.3.02.3.12.3.2Apr2.4.02.4.12.4.22.4.32.4.42.4.5May2.4.72.4.82.4.9Jun2.4.102.4.112.4.122.4.132.5.0OCR 4OCR 42.5.12.5.2Leanstral 1.5Leanstral 1.52.6.0Robostral NavigateRobostral NavigateJul2.7.02.7.12.7.22.8.02.9.02.9.1ShieldstralShieldstralAug2.9.22.9.32.9.42.10.0Sept2.10.13.0.03.03.1.03.2.0Large 4Large 4OctTimestralTimestralVoxtral 3Voxtral 3
One column per week of 2026. Above the line: releases of mistralai/client-python on GitHub, majors in red. Below it: launch posts on mistral.ai/news, named.

Out in 3.0.0

client=
must be an httpx2 client
tools: code_interpreter
use Conversations or an agent
tools: web_search
use Conversations or an agent
tools: web_search_premium
use Conversations or an agent
backend koyeb
renamed mistral_cloud
cpu_* / memory_* on deployments
removed

mistralai, Mistral's Python SDK, released 3.0.0 on 28 September 2026, six and a half months after 2.0.0. It moves from httpx to httpx2, so a client passed as client= or async_client= must now be an httpx2 client. Chat and agents completions no longer accept the code_interpreter, web_search and web_search_premium tools; those go through the Conversations API or an agent now. Workflow deployments lose their CPU and memory fields, and the koyeb backend is renamed mistral_cloud.

Three model launches followed in October: Mistral Large 4 on 6 October; Timestral on 15 October, a probabilistic forecasting model in four sizes from 5M to 209M parameters; and Voxtral 3 on 20 October, with new Transcribe and Speech models and a voice agent SDK built on LiveKit. SDK 3.1.0 and 3.2.0 came out on 6 and 9 October and mostly reshape the beta surface, removing 80 beta names between them.

As of 24 October 2026. mistralai 3.0.0 release notes.

Latest releases

  1. 3.2.0VoiceCreateRequest.sample_audio becomes optional; 64 beta names removed.
  2. 3.1.0Workflows.runs becomes WorkflowsRuns; Records removed from beta.
  3. 3.0.0httpx2; built-in tools leave chat and agents completions.
  4. 2.10.1Adds managed indexes; pipeline config fields reshaped.
  5. 2.10.0Adds service accounts; campaigns and explorer endpoints removed.
  6. 2.9.4Regenerated from the OpenAPI document.

From @MistralAI on X

Mistral AI@MistralAI

Introducing Voxtral 3: state-of-the-art Transcribe and Speech models powering Mistral鈥檚 new voice agents API. 馃У

On X, 20 Oct 2026 16:00:12 UTC

Voxtral 3 is the most recent of the three October launches.

Mistral AI@MistralAI

Voxtral 3 also introduces a voice agent SDK, built on LiveKit.
Transcribe listens, a reasoning model thinks and calls tools, and Speech replies, streamed as one pipeline for natural, intelligent conversation.

On X, 20 Oct 2026 16:00:15 UTC

The voice agent SDK chains Transcribe, a reasoning model with tool calls, and Speech.