Mistral has released 20+ Model Context Protocol (MCP) connectors for Le Chat, enabling integration with enterprise tools including Databricks, Snowflake, GitHub, Atlassian, Asana, Stripe, and Zapier. The company also introduced Memories, a cross-conversation context system that retains user preferences and facts. Both features are available on Le Chat's free plan.
Mistral AI reports that fine-tuning its Pixtral-12B vision model on satellite imagery increased classification accuracy from 56% to 91% on the Aerial Image Dataset. The company used LoRA (Low-Rank Adaptation) to train on 8,000 samples for under $10, reducing hallucinations from 5% to 0.1%.
Mistral AI released Codestral 25.08, showing 30% more accepted code completions and 10% higher retention rates. The company also shipped Devstral Small, a 24B-parameter agentic coding model scoring 53.6% on SWE-Bench Verified, alongside new embedding and IDE integration tools aimed at enterprise deployment.
Mistral AI released Voxtral, a family of open-source speech understanding models available in 24B and 3B parameter variants under Apache 2.0 license. The models support up to 32K token context (30 minutes of audio for transcription, 40 minutes for understanding) and are priced at $0.001 per minute via API—less than half the cost of comparable proprietary systems according to Mistral.
Mistral AI has released two specialized coding models: Devstral Medium, achieving 61.6% on SWE-Bench Verified, and Devstral Small 1.1, scoring 53.6% and released under Apache 2.0 license. The company claims Devstral Medium surpasses Gemini 2.5 Pro and GPT-4.1 at a quarter of the price.
Mistral AI has launched Mistral Compute, an AI infrastructure service offering private, integrated stacks including GPUs, orchestration, and APIs. The service will provide access to tens of thousands of NVIDIA GPUs, targeting European, Middle Eastern, and Asian customers seeking alternatives to US or China-based cloud providers.
Mistral has launched AI Studio, a development platform for building AI applications, alongside two new models: Mistral 3, its latest flagship, and Mistral Small 4, a cost-efficient alternative. The releases include new pricing tiers and API access through the unified platform.
Mistral AI has released Magistral, its first reasoning model line, in two versions: Magistral Small (24B parameters, Apache 2.0) and Magistral Medium (enterprise). Magistral Medium scored 73.6% on AIME2024 (90% with majority voting at 64 samples), while the open-source Small version achieved 70.7% (83.3% with voting).
Mistral AI has launched Le Chat Enterprise, powered by its new Mistral Medium 3 model. The platform includes enterprise search across Google Drive, Sharepoint, OneDrive, Gmail and Google Calendar, no-code agent builders, custom data connectors, and hybrid deployment options including self-hosted and cloud.
Mistral AI launched Mistral Medium 3 on May 7, 2025, priced at $0.4 per million input tokens and $2 per million output tokens. The company claims the model performs at or above 90% of Claude Sonnet 3.7 on benchmarks while being significantly less expensive, and surpasses Llama 4 Maverick and Cohere Command A.
Mistral AI has released an OCR API priced at $1 per 1,000 pages with batch inference costs approximately half that rate. The company claims 94.89% overall accuracy on internal benchmarks, ahead of GPT-4o (89.77%), Gemini 2.0 Flash (88.69%), and Azure OCR (89.52%). The model processes up to 2,000 pages per minute on a single node.
Mistral AI has released Saba, a 24B-parameter model trained specifically for Arabic and South Asian languages including Tamil. The model runs on single-GPU systems at over 150 tokens per second and is available via API or for on-premises deployment.
Mistral AI announced a partnership with Agence France-Presse (AFP) to integrate the news agency's newswire into Le Chat. The integration will provide access to 2,300 daily stories in French, English, Spanish, Portuguese, German, and Arabic from AFP's network of 1,700 journalists.
Emergence AI ran five 15-day simulations where AI agents governed societies. Claude Sonnet 4.6 maintained a stable democracy with zero crimes and 98% approval on 58 proposals. Grok 4.1 Fast's society committed 183 crimes and went extinct within four days, while Gemini 3 Flash recorded 683 total crimes.
NVIDIA released LocateAnything-3B, a 3-billion parameter vision-language model that predicts bounding boxes in parallel rather than token-by-token, achieving up to 2.5× higher throughput compared to autoregressive approaches. The model, trained on 12M images with 138M+ queries and 785M bounding boxes, supports object detection, GUI element grounding, and robotics perception.
Google has added two new voices to Gemini Live—Flare and Glow—replacing the previous Nova and Lyra options. The voice picker interface has been redesigned from a carousel to a list format, while voice descriptions have been removed entirely.
Amazon Web Services released Amazon Bedrock Data Automation (BDA), a foundation model-powered service designed to extract and validate structured data from financial documents. The service uses custom blueprints to process bank statements, W-2 tax forms, 1099-B forms, and vendor contracts, offering what AWS claims is industry-leading accuracy at lower cost than using foundation models directly.
Google is rolling out camera-based automations for Gemini for Home, allowing smart home cameras to trigger routines based on visual events like package deliveries or glass breaking. The update includes improved multi-request handling and reduced error rates for the AI assistant that replaced Google Assistant on smart home devices.
AWS has deployed Amazon Bedrock AgentCore to orchestrate more than 20 specialized AI agents across its global sales organization through a unified interface called Field Advisor. The system has processed over 120,000 prompts since launch, delivering a 41% latency reduction compared to previous infrastructure and saving large-scale sales reps up to 2 hours per week on CRM tasks.
Artificial Analysis and IBM Research have released ITBench-AA, the first benchmark evaluating AI models on enterprise Site Reliability Engineering tasks. Claude Opus 4.7 leads at 47%, followed by GPT-5.5 at 46% and Qwen3.7 Max at 42%—all frontier models score below 50% on Kubernetes incident response tasks requiring root-cause diagnosis across complex infrastructure.