Mistral AI News

Mistral finalizes trillion-parameter open-weight Le Chonk; Mensch warns of AI export controls

Mistral says Le Chonk (Mistral Large 4) is the best open-weight model for coding and security work, with three reasoning modes. Co-founder Arthur Mensch warned Europe could lose access to frontier models.

2026-11-04

Mistral releases open weights for trillion-parameter Large 4 'Le Chonk' aimed at coding and cyber defense

Mistral published weights for Mistral Large 4, a 1-trillion-parameter multimodal mixture-of-experts model with 52B active parameters per token. It leads open models on Artificial Analysis' Coding Agent Index, matching GLM-5.3 while using 47% fewer tokens, and scores 41% on Terminal-Bench 4.0. Orange Cyberdefense is the first European MSSP design partner for a cyber-optimized variant.

2026-11-03

Mistral opens weights of trillion-parameter Large 4 despite its own cyberattack-uplift warning

Mistral publicly released open weights for Mistral Large 4 ('Le Chonk'), weeks after making it available through its API. It is a roughly 1T-parameter Mixture-of-Experts model with 52B active parameters and a 1M-token context window. It scores 79% on CyberGym-E2E and 41% on Terminal-Bench 4.0 while using 47% fewer tokens than competitors. Mistral itself warned that the model could help attackers carry out real cyberattacks.

2026-11-02

Mistral releases open weights for 1.05T-parameter Large 4 'Le Chonk', pitching defensive cyber

Mistral released open weights for Mistral Large 4, nicknamed 'Le Chonk', a 1.05-trillion-parameter mixture-of-experts model activating about 52B parameters per token over a 1M-token context. Building on an October 6 preview, it targets agentic coding and defensive cybersecurity, with a reported 96.6% refusal rate on malicious cyber requests. The release drew 1,706 HN points.

2026-11-01

Mistral releases open weights for trillion-parameter Large 4 'Le Chonk'

Mistral made Mistral Large 4 generally available as open weights on Hugging Face under a modified MIT license, after an October preview and further reinforcement learning. The 1T-parameter model (~52B active) scores 41% on Terminal-Bench 4.0, tying GLM-5.3 while using 47% fewer tokens, and API pricing is $0.70/$2.10 per million input/output tokens. Mistral pitches it for sovereign cyber defense, trained in its own European data centers.

2026-10-31

Mistral's Mensch bets on a self-owned GPU fleet, with a cyber-focused open model due within a week

Mistral runs its own GPU fleet in European data centers rather than renting from cloud providers. Its next open-source release reportedly carries state-of-the-art cyber capabilities. The company is preparing for an IPO within a few years.

2026-10-29

Mistral rebuilds itself as a full AI stack around Large 4, with Vibe, Studio, Forge and Compute

Mistral launched Vibe (a work and coding agent), Studio, Forge and Compute alongside Large 4. Large 4 is a 1.05T-parameter model in public preview since October 6, with open weights expected by end of October. The company also added document AI, speech and industrial engineering offerings to compete with OpenAI and Google as a full-stack provider.

2026-10-26

Mistral and NVIDIA pair 1T-parameter Mistral Large 4 with CUDA-X for engineering simulation

Mistral is collaborating with NVIDIA on AI-native industrial engineering. The partnership pairs Mistral Large 4 (1T parameters, 49B active) with CUDA-X, Warp and Omniverse for CAE and EDA workflows. Mistral bills Large 4 as one of the strongest open-weight models available.

2026-10-25

Mistral previews Large 4 'Le Chonk', a ~1T-parameter open-weight model trained in Europe

Mistral Large 4 has about 1 trillion total and 49 billion active parameters. It was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-run European data centers. Mistral claims it ranks among the strongest open-weight systems, with gains in coding, agentic workflows and multimodal tasks.

2026-10-24

Mistral and NVIDIA partner on AI for industrial engineering with open CAE benchmarks

Mistral models will be combined with NVIDIA accelerated computing, physics tools and Omniverse for closed-loop AI-driven engineering design. The partners will publish open benchmarks and datasets for physics AI, plus evaluations for agentic AI in computer-aided engineering (CAE) and electronic design automation (EDA).

2026-10-23

Mistral launches Voxtral 3 speech models and a real-time voice agent SDK

Voxtral 3 includes Small Transcribe 3 (50+ languages), Mini Transcribe Realtime 3, and Speech 3 TTS (35+ languages), priced from $0.0035 per minute to $13 per 1M characters. A LiveKit-based voice agent SDK combines them into a single streaming pipeline for real-time spoken conversation.

2026-10-22

Mistral Large 4 'Le Chonk' preview priced at $1.36/M input; open weights due October 31

Mistral Large 4 is a 1.05T-parameter MoE model with 49B active parameters. It is in public API preview at $1.36/M input and $4.18/M output tokens, with full open weights due October 31. It scored 38 on Artificial Analysis's index, trailing DeepSeek V4.1 Flash.

2026-10-17

Ecosia drops Mistral for Qwen, GLM and Kimi, citing a 'one-year quality gap'

Berlin-based search engine Ecosia replaced Mistral, which it had adopted five months earlier, with Chinese open-weight models: Alibaba's Qwen, Zhipu's GLM and Moonshot's Kimi. CEO Christian Kroll cited quality lag and server overloads, and reported lower costs on the new models.

2026-10-11

Mistral Large 4 'Le Chonk': 1T-parameter open-weight model in preview, weights due October 27

Mistral Large 4 is a natively multimodal 1T-parameter mixture-of-experts model activating 52B parameters per token, trained in European data centers on about 3,800-4,000 NVIDIA Grace Blackwell GPUs. It scores 63% on Deep SWE 1.1, and Mistral calls it the best open-weight model outside China. The API preview is live now, with open weights due October 27 after red-teaming.

2026-10-09

Mistral closes €3B Series D above €21B valuation and pledges 1GW of European compute

Mistral raised €3B at a valuation above €21B (about $24B) and committed to building one gigawatt of European compute by 2030. A new Munich hub focuses on Physics AI and Industrial AI, including BMW crash-simulation work. It is staffed by 30+ physicists from the acquired Emmi AI.

2026-10-05

Mistral reaches $24B valuation and ships Shieldstral 1.0 open-weight safety classifier

Mistral closed a $3.58 billion round that values it at $24B, a European record. It also released Shieldstral 1.0, an open-weight multimodal safety classifier built on Ministral-3-3B that runs on a single consumer GPU. Shieldstral takes a content policy written in plain language and returns a calibrated safety score against it.

2026-10-04

Mistral ships Shieldstral 1.0, an open-weight policy classifier, after reaching a $24B valuation

Mistral released Shieldstral 1.0, a multimodal open-weight safety classifier that scores content against plain-language policies on the fly. It is built on Ministral-3-3B and runs on a single consumer GPU, removing the need for separate per-policy moderation models. It follows a $3.58B funding round that valued Mistral at $24B.

2026-10-03

Mistral opens Munich industrial-AI hub as Mensch says US rivals use safety as a distraction

Mistral opened a Munich hub for Physics AI and Industrial AI research teams, working with BMW on crash simulations, Siemens Energy and TU Munich. CEO Arthur Mensch told CNBC that US labs use safety as a distraction, and claimed Mistral's next-generation model will close the gap with them "very significantly." Mistral recently raised €3B at a €21B valuation.

2026-10-01

Mistral CEO Arthur Mensch calls U.S. safety debate cover for 'negligence', vows not to slow down

Mistral CEO Arthur Mensch told CNBC the company won't slow down and accused American rivals of using the safety debate as cover for 'negligence.' He said Mistral's next-generation model will 'very significantly' narrow the gap with U.S. labs. The comments come as Mistral opens a Munich hub for physics and industrial AI following its €3B raise.

2026-09-30

Mistral Makes Chinese GLM 5.3 the Default in Vibe Code, Sparking Independence Debate

Mistral integrated the Chinese-origin GLM 5.3 model from Z.ai as the default in its Vibe Code assistant, featuring a 1-million-token context window, 100+ tokens/second throughput, a 99.5% uptime SLA, and 90% cached-input pricing cuts. The move sparked debate given Mistral previously defaulted to its own Medium 3.5 and Large 3 models.

2026-09-28

Mistral Raises €3B Led by Samsung, Promises New Flagship 'Within Weeks'

Mistral AI raised €3 billion led by Samsung Electronics — the largest private tech fundraise in European history — and CEO Arthur Mensch announced a new large flagship launching within weeks, with visibility into Mistral Large 4, 5, and 6 over 12 months using roughly 20x more training compute. Mensch also accused American labs of using safety fears for regulatory capture.

2026-09-27

Mistral raises €3 billion led by Samsung in Europe's largest private tech round

Paris-based Mistral AI raised €3 billion led by Samsung Electronics — the largest private tech fundraising in Europe to date. The company is doubling down on a specialist-model strategy and controlling both infrastructure and intelligence, including its OCR 4.1 model that powers its Vibe document library.

2026-09-26

Mistral raises €3 billion in Europe's largest private tech round, led by Samsung

Paris-based Mistral AI raised €3 billion in a Series D — the largest private tech fundraising in European history — led by Samsung Electronics, pushing its valuation above €21 billion. Mistral is deepening cooperation with the European Space Agency and pursuing specialist models like OCR 4.1, which transcribes handwriting and powers its Vibe document library. Its VP of Engineering argued AI productivity won't come from models alone.

2026-09-25

Mistral raises record €3B Samsung-led round at €21B valuation and acquires Pimento

Mistral closed a €3 billion Series D led by Samsung — Europe's largest tech funding round — valuing the company at over €21 billion, and acquired French startup Pimento to add AI advertising capabilities to its Vibe conversational assistant. Separately, Mistral denied a fresh security breach after code resembling its source code surfaced on dark-web forums, saying it found no evidence of new unauthorized access. A Counterpoint report also found Mistral holds a 19% share of adapted sovereign LLMs.

2026-09-23

Mistral Closes €3B Series D Led by Samsung at €21B Valuation

Mistral announced Europe's largest tech equity round — a €3 billion Series D led by Samsung Electronics, with EQT and PSG Equity — valuing the company above €21 billion. Funds target its own data centers and sovereign inference infrastructure, and Mistral publicly rejected US calls for an AI development slowdown.

2026-09-21

Mistral Partners with Mozilla to Power Firefox's AI Browsing Assistant

Mistral AI partnered with Mozilla to integrate its European models into Firefox Smart Window (beta), Mozilla's AI browsing assistant, initially for users in France and North America with UK and Germany planned. The tie-up emphasizes private, multilingual AI browsing built on open technology and regionally fine-tuned models.

2026-09-21

Mistral launches $115M TotalEnergies oil-and-gas AI program

Mistral and TotalEnergies launched a three-year, $115M+ program to build frontier AI models for oil-and-gas exploration, reservoir characterization, and field development, combining nearly 10 petaflops of compute and data with a century of geoscience expertise. The deal marks Mistral's aggressive vertical monetization strategy.

2026-09-18

Mistral powers Firefox's Smart Window AI browsing in France and North America

Mistral announced a partnership with Mozilla to integrate its models into Firefox Smart Window (beta), Mozilla's AI browsing assistant, delivering private, multilingual AI browsing to users in France and North America, with the UK and Germany to follow. The tie-up emphasizes open-source and locally optimized AI.

2026-09-18

Mistral Strikes €100M+ TotalEnergies Oil-and-Gas AI Program

Mistral AI announced a three-year, €100M+ ($115M+) partnership with French oil major TotalEnergies to develop proprietary frontier models for oil and gas exploration, reservoir characterization, and field development through a joint scientific laboratory. The deal follows Mistral's record €3 billion Series D at a €21B+ valuation led by Samsung.

2026-09-16

Mistral closes record €3B Series D led by Samsung at €21B+ valuation

Mistral closed a €3 billion Series D at a post-money valuation exceeding €21 billion ($24.4B) — Europe's largest tech equity round — led by Samsung with BlackRock, Advent, and PSG Equity joining. It also partnered with Cloudera to run frontier models across hybrid, sovereign, and fully air-gapped environments serving 30 exabytes of enterprise data.

2026-09-14

Mistral raises record €3B Series D at €21B+ valuation led by Samsung, partners with Cloudera

Mistral AI closed a €3 billion Series D at a €21 billion-plus valuation — Europe's largest-ever equity round — led by Samsung with EQT and PSG Equity. A parallel Cloudera partnership will deploy sovereign Mistral models across 30+ exabytes of enterprise data in hybrid and air-gapped environments. Samsung also partnered separately with Mistral on on-premises AI for semiconductor manufacturing.

2026-09-13

Mistral closes record €3B Series D led by Samsung at €21B+ valuation

Mistral AI raised €3 billion ($3.58B) in a Series D at a post-money valuation above €21 billion — the largest equity round ever by a European tech company. Samsung Electronics led, joined by co-leads EQT's Scaleup Europe Fund and existing investor PSG Equity, with NVIDIA, ASML, Salesforce Ventures, a16z and Luxembourg (€10M) participating. Funds target frontier research, compute and international expansion.

2026-09-12

Mistral partners with Cloudera to run sovereign models on governed enterprise data

Cloudera announced a nine-figure strategic partnership with Mistral to integrate the French company's frontier open-weight models directly into its data platform, letting enterprises run AI on governed data without moving it externally. The deal targets sovereign enterprise AI workloads across hybrid data environments, complementing Mistral's push into controllable, auditable production systems.

2026-09-12

Mistral raises €3B led by Samsung at €21B valuation — Europe's largest tech round

Mistral closed a €3 billion ($3.5B) Series D led by Samsung Electronics at a €21 billion post-money valuation, nearly doubling its prior €11.7B mark and marking Europe's largest-ever tech equity round. Co-leads include the EU-backed Scaleup Europe Fund and PSG Equity, with Advent and BlackRock participating. Samsung and Mistral also announced an AI chip design and manufacturing partnership.

2026-09-11

Mistral partners with Cloudera to bring sovereign AI to enterprise data

Mistral announced a nine-figure strategic partnership with Cloudera to integrate its frontier models into Cloudera's hybrid data platform. The collaboration lets enterprises deploy and customize AI agents across environments while keeping data fully governed and in-place, with no external data movement. Mistral also highlighted new Agentic Search for complex document navigation.

2026-09-11

French minister warns Europe is 'doomed' relying on Mistral for AI sovereignty

France's Finance Minister Roland Lescure warned that European AI sovereignty cannot depend on a single startup like Mistral. Valued at nearly €12 billion, Mistral is building its first French data center and has a commercial agreement with Microsoft. Concerns include the risk of the U.S. restricting Europe's access to top models. Mistral models were also added to Oracle's OCI Enterprise AI Model Import catalog.

2026-09-06

Mistral signs strategic AI partnership with Côte d'Ivoire

Mistral signed a strategic partnership with Côte d'Ivoire (via MTNIT) to structure the country's national AI ecosystem, including a full audit, an operational roadmap, and pilot projects across government, health, education, and agriculture. Separately, a French minister warned against over-relying on Mistral, noting US frontier models lead and Chinese models are close behind. Mistral is valued at nearly €12 billion.

2026-09-05

French minister warns Europe against over-relying on Mistral for AI

French minister Lescure warned that if European AI is reduced to Mistral alone, 'we are lost,' noting US frontier models lead and Chinese models are close behind. Valued at nearly €12 billion, Mistral is often called Europe's only top-tier model maker but still trails leading American and Chinese competitors. The European Commission has tasked an Italian-German consortium — excluding Mistral — with developing a state-of-the-art model on EU public supercomputers.

2026-09-04

Mistral partners with Côte d'Ivoire and Saudi Arabia's Humain on sovereign AI

Mistral partnered with Côte d'Ivoire to develop AI for national priorities and struck a collaboration with Saudi Arabia's Humain to build sovereign AI infrastructure, model development and regional deployment. The deals extend Mistral's sovereign-AI positioning across Africa and the Middle East.

2026-09-03

France names Mistral sovereign AI provider for government cybersecurity, excludes OpenAI

French Budget Minister David Amiel announced the government will hire sovereign AI providers such as Mistral — explicitly excluding OpenAI — to test public services for vulnerabilities after a DGFiP breach exposed ~700,000 taxpayers. Mistral also released Agentic Search, a retrieval layer reducing token use and latency.

2026-09-02

Mistral and HUMAIN partner on sovereign AI in Saudi Arabia, focused on cybersecurity and voice

Mistral and HUMAIN announced a collaboration worth hundreds of millions of euros to develop and localize advanced AI models in Saudi Arabia, with initial focus on cybersecurity and voice. The companies also plan to build frontier models that perform strongly in Arabic to support the broader region, deepening Mistral's sovereign-AI strategy.

2026-08-30

Mistral and HUMAIN Launch Sovereign-AI Partnership for Saudi Arabia

Mistral and Saudi Arabia's HUMAIN announced a multi-hundred-million-Euro strategic collaboration to advance sovereign AI across the Middle East, spanning AI infrastructure, model development, cybersecurity, and voice, with a focus on frontier models strong in Arabic.

2026-08-25

Mistral Introduces Agentic Search, Tripling Accuracy on Financial Filings

Mistral introduced Agentic Search, a retrieval layer that helps AI systems navigate, read, and verify complex documents with better accuracy and fewer turns. It reports up to 3x accuracy gains on financial filings (26.7% to 86% on FinanceBench), one-third lower token use, and up to 39.6% lower p90 latency.

2026-08-23

France Taps Mistral AI for Cybersecurity, Formally Excludes OpenAI

France called on Mistral AI to detect vulnerabilities in its public services while formally excluding OpenAI, relying on the 'Our AI' ecosystem and SecNumCloud-certified data centers for digital sovereignty. The move follows a breach exposing data of nearly 700,000 taxpayers via a compromised VPN. Mistral also recently acquired Emmi AI to expand into physics AI.

2026-08-20

Mistral expands sovereign AI with regional inference and a European compute coalition

Mistral AI is expanding its sovereign-AI infrastructure with regional inference endpoints, a new Priority Tier for mission-critical workloads, and support for third-party open models starting with Z.ai's GLM-5.2. It is also forming a coalition to secure long-term European compute, aiming for up to 1 GW by 2030, as France announced plans to lean on domestic AI firms like Mistral amid IPO speculation.

2026-08-19

Mistral signs five European anchor customers for 1GW sovereign compute plan, opens platform to third-party models

Mistral secured multi-year commitments from ASML, CMA CGM, Amadeus, Capgemini and Caisse des Dépôts to buy European Compute Units under a 1GW sovereign infrastructure plan (200MW by end-2027, scaling to 1GW by 2030). It also launched a paid Priority Tier with a 99.5% uptime SLA and, for the first time, will host third-party open models — starting with Chinese lab Z.ai's GLM.

2026-08-18

Mistral pivots to infrastructure, hosting external models starting with Z.ai's GLM-5.2

Mistral announced it will host external open models alongside its own, launching China's Z.ai GLM-5.2 first, so customers can switch models without separate cloud providers. The move is part of a three-step AI sovereignty push that also makes Regional Endpoints generally available and introduces a Priority Tier with an SLA. Mistral also formed a coalition committing 200MW of European compute by end-2027 and 1GW by 2030.

2026-08-17

Mistral opens platform to third-party open models and plans 1GW of EU compute

Mistral began hosting third-party open-weight models on its own infrastructure for the first time, starting with Z.ai's GLM-5.2, alongside new in-region inference endpoints for Europe and the US and a Priority Tier with uptime SLAs. The Paris-based company also revealed plans to build a European enterprise compute coalition committing to 200MW by end-2027 and 1 gigawatt by 2030 via European Compute Units.

2026-08-15

Mistral releases Shieldstral 1.0 3B, a policy-adaptive multimodal safety classifier

Mistral released Shieldstral 1.0 3B, an open-weights, policy-adaptive multimodal safety classifier built on Ministral-3-3B-Base with Pixtral vision. It reframes moderation as a plain-language yes/no question specified at inference, scoring 84.9% F1 on text safety and 83.8% on multimodal safety, matching models seven times larger while running on a single 16GB GPU under Apache 2.0.

2026-08-11

Mistral open-sources Shieldstral 3B multimodal safety classifier

Mistral released Shieldstral 1.0 3B, an Apache-2.0 open-weights policy-adaptive multimodal safety classifier built on Ministral-3-3B-Base with a native Pixtral vision encoder. It achieves 84.9% average F1 on text safety — matching GPT-OSS-Safeguard-20B — and 83.8% on multimodal safety, runs on a single 16GB GPU, supports 12 languages and adapts to new policies without retraining.

2026-08-10

Mistral launches Shieldstral, a 3B open-weight policy-aware multimodal safety classifier

Mistral released Shieldstral 1.0, a 3-billion-parameter open-weights safety classifier under Apache 2.0 that evaluates text and images against natural-language safety policies specified at inference time. It runs on a single 16GB GPU, supports 12 languages, achieves an 83.8% average image-safety benchmark, and matches guard models up to seven times larger without retraining when policies change.

2026-08-08

Mistral Small 4 Sets Sparse-Efficiency Benchmark: 119B Total, 6.5B Active Parameters

Mistral released Mistral Small 4 with 119 billion total but only 6.5 billion active parameters, running in roughly 142GB of GPU memory at FP4 precision — a new open-source sparse-activation efficiency benchmark. The company also acquired Austrian physics-AI group Emmi AI and released Voxtral TTS, an OCR 4 tool, and a Search Toolkit.

2026-08-04

Mistral nears €20B valuation, partners with Azure for European AI infrastructure

Mistral AI is in advanced talks at a €20 billion ($22 billion) valuation and locked in a multibillion-dollar deal with Microsoft, where Azure funds its European data-center expansion. Two new models — Medium 3.5 and OCR 4 — are now available on Microsoft Foundry, positioning Mistral as a sovereign European alternative.

2026-08-01

Mistral enters Physical AI with Robostral Navigate at €20B valuation

Mistral AI unveiled Robostral Navigate, its first robotics model, enabling autonomous navigation from a single camera and basic language prompts and achieving a 76.6% success rate — outperforming prior single-camera approaches by 9.7 points. The launch marks Mistral's move into Physical AI, following deals with Airbus and BMW, as the European lab targets a €20 billion valuation.

2026-07-31

Microsoft and Mistral expand partnership with multibillion-dollar European compute deal

Microsoft and Mistral announced a multibillion-dollar agreement to expand AI infrastructure across Europe, with Microsoft funding Mistral's Europe-based GPU buildout. Mistral Medium 3.5 and OCR 4 are now available in Microsoft Foundry, with Medium 3.5 integrated into Copilot Studio, and Azure customers can build on Mistral's French data centers for regulated, sovereign, and disconnected operations.

2026-07-24

Microsoft commits multibillion dollars to Mistral's European AI infrastructure

Microsoft and Mistral expanded their strategic partnership with a multibillion-dollar commitment to grow Mistral's GPU infrastructure in Europe on NVIDIA Vera Rubin hardware, positioning it as a non-U.S. cloud alternative for regulated industries. Mistral Medium 3.5 and OCR 4 are now available in Microsoft Foundry, with Medium 3.5 integrated into Copilot Studio and deployable across cloud, cloud-connected, and disconnected Azure Local environments.

2026-07-23

American open-source labs race to counter China's AI models

Forbes reports France's Mistral earned over $200M last year from open-weight models pitched to Western companies and governments on a 'not Chinese and not American' selling point, as U.S. startups like Poolside — releasing its open model Laguna — race to build homegrown alternatives to Chinese open models threatening American AI hegemony. NVIDIA publicly congratulated Poolside on Laguna S 2.1, an open-weight model that 'runs great locally.'

2026-07-23

Microsoft signs multibillion-dollar deal to fund Mistral's European AI expansion

Microsoft agreed to spend billions on Mistral's European computing infrastructure and expand distribution through Azure, in an expanded global strategic partnership. Mistral added its Medium 3.5 and OCR 4 models to Copilot Studio and Microsoft Foundry, with Azure Local now supporting Mistral's open models. The pitch is frontier AI that regulated European industries can control.

2026-07-22

Mistral launches Robostral Navigate, single-RGB-camera robot navigation model

Mistral AI released Robostral Navigate, an 8-billion-parameter model that lets robots autonomously navigate complex environments using only a single RGB camera and plain-language instructions. It achieved a 76.6% success rate on unseen R2R-CE benchmarks, outperforming multi-sensor approaches, and was trained entirely in-house on simulated data.

2026-07-16

Mistral launches Robostral Navigate, an 8B robot-navigation model using a single camera

Mistral unveiled Robostral Navigate, an 8-billion-parameter model that enables autonomous robot navigation from a single RGB camera and natural-language instructions — no depth sensors or LiDAR. Trained entirely in simulation across 400,000 trajectories and 6,000 virtual scenes, it hit a 76.6% success rate on unseen R2R-CE benchmarks, outperforming multi-sensor systems, as Mistral pushes into physical AI.

2026-07-13

Mistral enters physical AI with Robostral Navigate, an 8B robotics model

Mistral launched Robostral Navigate, an 8-billion-parameter model for 'embodied navigation' that lets robots move through complex environments using only a single standard RGB camera and plain-language instructions. It scores 76.6% on the R2R-CE benchmark with superior efficiency, marking Mistral's first robotics entry.

2026-07-12

Mistral releases Robostral Navigate, an 8B robotics navigation model

Mistral AI launched Robostral Navigate, an 8B-parameter model that lets robots autonomously execute natural-language tasks using only a single RGB camera, achieving state-of-the-art results on the R2R-CE benchmark. The hardware-agnostic model completes entire tasks from a single instruction, adapting to offices and outdoor environments without prior exposure, extending Mistral's physical-AI push after its May acquisition of Austria's Emmi AI.

2026-07-11

Mistral cast as Europe's sovereign OpenAI rival as ARR nears $1B and new open-weight model looms

Mistral AI is increasingly positioned as Europe's answer to OpenAI amid a push for sovereign tech that reduces reliance on US firms. TechCrunch reports Mistral is rumored to be raising ~$3.5B at a $23.15B valuation, has grown ARR from $20M to over $400M in a year (targeting $1B), follows a Palantir-style forward-deployed-engineer playbook, and plans a new open-weight model this summer.

2026-07-06

Mistral releases Leanstral 1.5, an Apache-2.0 Lean 4 code agent solving 587/672 PutnamBench problems

Mistral released Leanstral 1.5, an Apache-2.0-licensed Lean 4 code agent model for automated theorem proving under the banner 'Proof Abundance for All.' Part of the Mistral Small 4 family, it solves 587 of 672 PutnamBench problems and achieves state-of-the-art FATE-H (87%) and FATE-X (34%) algebra benchmarks, available via a free API and on Hugging Face.

2026-07-06

Mistral emerges as a symbol of European tech sovereignty, teasing a new open-weight model this summer

Mistral AI is drawing spotlight as a symbol of European technological independence amid calls to reduce reliance on US AI, following a Trump directive that led Anthropic to pull models offline. The French firm plans a new open-weight model this summer with July early access, claiming state-of-the-art voice, image and document handling.

2026-07-05

Mistral releases Leanstral 1.5 for automated theorem proving in Lean 4

Mistral AI released Leanstral 1.5, a specialized model for automated theorem proving and formal verification with the Lean 4 proof assistant. It uses a mixture-of-experts architecture with 119 billion parameters (6.5 billion active) and a 256k-token context length to handle long proof files, and is available free in Mistral's Labs tier.

2026-07-03

Mistral AI launches OCR 4 document intelligence model on Microsoft Foundry and AWS

Mistral AI released OCR 4, a document intelligence model that extracts structured data from unstructured documents across 170 languages, with paragraph-level bounding boxes, typed-block labels, and confidence scores to improve RAG pipelines. It is available via the Mistral API, Mistral Studio, Amazon SageMaker, and Microsoft Foundry at $4 per 1,000 pages for standard API access.

2026-07-02

Mistral launches OCR 4, turning unstructured documents into layered data

Mistral released OCR 4, a document intelligence model that returns fully structured representations rather than raw text — with paragraph-level bounding boxes, typed-block labels, and confidence scores across 170 languages. It processes PDF, DOC, and PPT directly and is available via API, Mistral Studio, Amazon SageMaker, and Microsoft Foundry for advanced RAG pipelines.

2026-07-01

Mistral ships OCR 4 with bounding boxes and confidence scores across 170 languages

Mistral released OCR 4, a document-understanding model that goes beyond text extraction to provide structured output with bounding boxes, block classification and inline confidence scores across 170 languages. It is self-hostable in a single container, integrates with RAG, agentic and enterprise-search pipelines, and is priced at $4 per 1,000 pages via API. It's distributed through the Mistral API, Amazon SageMaker and Microsoft Foundry, with Microsoft highlighting the launch as a partnership milestone.

2026-06-26

Airbus partners with Mistral AI for documentation and pilot assistance

Airbus has partnered with Mistral AI to target use cases including technical documentation and pilot assistance, reflecting growing enterprise adoption of European AI models in aerospace and the sovereign-AI push gaining momentum this week.

2026-06-21

Mistral's Ministral-3-14B multimodal model arrives in SageMaker JumpStart

AWS added Mistral AI's Ministral-3-14B-Instruct-2512 to Amazon SageMaker JumpStart. The compact 14B-parameter multimodal model is optimized for edge deployment, enabling AI assistants, agentic systems, and vision-language applications.

2026-06-20

Mistral Nears €3 Billion Raise at ~€20 Billion Valuation

French AI startup Mistral is in advanced talks to raise approximately €3 billion (~$3.47B) at a valuation of roughly €20 billion (~$23.15B), nearly doubling the €11.7 billion it secured in last September's Series C and cementing it as Europe's leading AI player.

2026-06-15

Mistral raises ~€3B at a ~€20B valuation and launches industrial AI stack

Mistral is in advanced talks to raise around €3 billion at a roughly €20 billion valuation — nearly doubling its €11.7 billion September 2025 valuation in about eight months. The French startup also launched a dedicated industrial AI stack and signed multi-year contracts with Airbus, BMW, and EDF, targeting aerospace, automotive, and nuclear.

2026-06-14

Mistral in talks to raise ~€3B at a ~€20B valuation

France's Mistral AI is in advanced discussions to raise roughly €3 billion (about $3.5B) at a ~€20 billion valuation, nearly doubling its €11.7 billion September 2025 figure in eight months. The round aims to fund Europe's frontier AI lab in the compute race, and Mistral has reportedly discussed a cybersecurity-focused model to rival Anthropic's Mythos.

2026-06-13

Mistral launches Vibe unified agent, Mistral Medium 3.5, and Forge enterprise system

Mistral rebranded Le Chat as Vibe, a unified AI agent for work and code with dedicated Work and Code modes, a new VS Code extension, and CLI updates. The capabilities are powered by the new Mistral Medium 3.5 model, and Mistral also introduced Forge, letting enterprises build frontier-grade models grounded in proprietary knowledge.

2026-06-12

Mistral launches Medium 3.5 flagship and 'Vibe' agent with VS Code integration

Mistral AI launched Mistral Medium 3.5, a 128B dense flagship model unifying instruction-following, reasoning and coding that can be self-hosted on as few as four GPUs. It also introduced 'Vibe,' an AI agent for work and code with a Work Mode for long-running tasks, a Code Mode for remote coding, and a new VS Code extension.

2026-06-07

Mistral pitches 'good enough' AI and European sovereignty to Wall Street

Europe's biggest AI lab, Mistral, is courting US investors and customers with a bet that users don't need the smartest model, leaning on open-weight releases and French/European AI sovereignty. The company has raised $3 billion total at an €11.7 billion valuation with ~$400 million in annual revenue and a $1 billion revenue target for end of 2026.

2026-06-05

Mistral Medium 3.5 lands in Microsoft Copilot Studio

Mistral AI's Medium 3.5 model is now available in Microsoft Copilot Studio, giving users a new language model to integrate into workflows and offering European customers a data-sovereignty-friendly alternative within the Microsoft ecosystem.

2026-06-04

Mistral announces 10 MW inference facility in Les Ulis amid European sovereign-AI push

Mistral AI announced a new 10 MW facility in Les Ulis, Essonne, dedicated to inference operations and scheduled to open Q3 2026, addressing compute supply-chain risk by giving Mistral direct control over capacity. The move, unveiled around its Paris summit, supports a strategy to build European AI infrastructure independent of US and Chinese cloud providers.

2026-06-02

Mistral launches Vibe consumer agent and Industrial Engineering stack on new 128B Medium 3.5

Mistral AI rebranded its consumer assistant to 'Mistral Vibe' (with Work Mode for long tasks and Code Mode for remote coding) and launched 'Mistral for Industrial Engineering,' combining LLMs with Emmi AI physics simulation. Both run on the new flagship Mistral Medium 3.5 — a 128B dense model that self-hosts on as few as 4 GPUs. Mistral also announced a 10MW inference data center in Les Ulis and Microsoft Copilot Studio integration.

2026-05-30

Mistral Pivots to Physical AI with Airbus, BMW Partnerships and New French Data Center

Mistral announced customer partnerships with Airbus and BMW for advanced manufacturing and physical AI, plus a new French data center. CEO Arthur Mensch is positioning Mistral beyond LLMs into industrial and robotics applications.

2026-05-29

Mistral expands Harvey AI partnership to push into legal workflows

Paris-based Mistral AI expanded its partnership with legal-tech startup Harvey AI, integrating its models into Harvey's platform across 1,500+ organizations in 60+ countries. CRO Marjorie Janiewicz said Mistral has prioritized financial services and regulated industries since founding, leveraging European data-sovereignty positioning to attack a sector where Anthropic has been aggressive.

2026-05-27

Mistral CEO: Europe has 2 years to escape US AI dependence; acquires Emmi physical-AI startup

Mistral CEO Arthur Mensch warned Europe has a narrow two-year window to build sovereign AI infrastructure before US hyperscalers lock in permanent dominance. Mistral also acquired Viennese physical-AI startup Emmi to accelerate its industrial-AI push.

2026-05-24

Trump postpones AI security executive order amid industry pushback

The Trump administration postponed a planned executive order on AI security that would have formalized pre-release testing access between AI labs and US/UK regulators. A former federal official who saw the latest draft said the postponement followed industry concerns, even though the order largely codified existing voluntary cooperation.

2026-05-22

Mistral pitches sovereign cyber AI to European banks as Anthropic Mythos alternative; Medium 3.5 Vision ships

Mistral AI is in talks with European banks about a cybersecurity-focused AI model positioned as a sovereign, lower-risk alternative to Anthropic's Mythos — which European financial institutions largely cannot access. CEO Arthur Mensch separately told French Parliament that 'engineers at Mistral no longer write a single line of code,' and warned French authorities against letting US models scan military codebases. Mistral also released Medium 3.5 Vision, a 128B-parameter model claimed to match systems 6x its size.

2026-05-18

Mistral CEO warns France against letting Anthropic's Mythos scan military code

Arthur Mensch publicly warned France against letting Anthropic's Mythos scan military code bases, citing growing European cybersecurity dependency on U.S. AI. He confirmed Mistral is targeting an IPO and ruled out a sale. Mistral is also developing a cybersecurity-focused AI model for European banks as a domestic alternative to Mythos.

2026-05-17

Mistral pitches sovereign cybersecurity model to European banks

French AI startup Mistral is in discussions with European banks about a cybersecurity-focused AI model positioned as a lower-risk domestic alternative to Anthropic's restricted-access Mythos. The tool would help institutions identify vulnerabilities at speed and scale, and Mistral is preparing a standardized version for broader rollout.

2026-05-17

Mistral deploys generative AI to 100,000 Caisse des Dépôts agents; preps cybersecurity model for EU banks

Mistral landed a flagship sovereign-AI deployment, rolling out generative AI tools to 100,000 agents at France's Caisse des Dépôts. It is also developing a cybersecurity AI model for European banks lacking Anthropic Mythos access, addressing concerns about reliance on non-EU technology providers.

2026-05-16

Mistral building cybersecurity AI for European banks shut out of Mythos

French AI startup Mistral is in talks with European banks to deploy a new cybersecurity-focused model as an alternative to Anthropic's restricted-access Mythos. The pitch is sovereign AI for vulnerability detection — at a moment when AISI just confirmed Mythos shattered cyber benchmarks.

2026-05-15

Mistral builds cybersecurity AI for European banks shut out of Mythos

French AI startup Mistral is in talks with European banks to deploy a cybersecurity-focused AI model as an alternative to Anthropic's restricted-access Mythos, which has demonstrated unprecedented offensive-cyber capability per UK AISI evaluations.

2026-05-14

Mistral Medium 3.5 launches with 256k context; Vibe remote coding agents debut

Mistral AI released Mistral Medium 3.5, a 128B dense model with 256k context combining instruction-following, reasoning, and coding, under a modified MIT license with open weights. Alongside it, the company introduced 'Vibe' remote coding agents — asynchronous cloud agents launchable from CLI or Le Chat that persist while users step away.

2026-05-11

Mistral Medium 3.5 ships open-weight 128B with remote coding agents and Le Chat Work Mode

Mistral released Medium 3.5, a 128B-parameter dense model unifying instruction-following, reasoning and coding under a modified MIT license, self-hostable on as few as four GPUs. It powers new remote coding agents in Mistral Vibe (running asynchronously in the cloud) and a Work Mode in Le Chat for multi-step workflows across connected tools. The model is live on NVIDIA NIM and Hugging Face.

2026-05-07

Mistral deploys AI agents to 100,000 French public-sector workers

Mistral signed a landmark deal with La Caisse des Dépôts to roll out generative AI agents to 100,000 public-sector workers — a flagship 'sovereign AI' win in Europe. Mistral also shipped Medium 3.5 with an upgraded remote coding agent in Vibe and a new Work mode in Le Chat.

2026-05-06

Voxtral TTS uses hybrid autoregressive + flow-matching for multilingual voice cloning

Mistral unveiled Voxtral, a TTS system that pairs an autoregressive backbone with flow-matching synthesis to close the 'expressivity gap' in multilingual voice cloning. The architecture aims to preserve speaker identity, emotion, and rhythm beyond the first few seconds — a known failure mode for current systems.

2026-05-06

Mistral Medium 3.5 hits 77.6% SWE-Bench Verified, ships open weights

Mistral Medium 3.5 — a 128B dense model with 256k context — scored 77.6% on SWE-Bench Verified and is now available as open weights on Hugging Face. It powers Mistral's new Vibe remote agents and Le Chat.

2026-05-04

Mistral Workflows enters public preview; Medium 3.5 ships with vibe remote agents

Mistral launched Workflows in public preview, pivoting from model-centric competition to enterprise agent orchestration, and released Medium 3.5 with vibe remote agents (450 HN points). The company also secured $830M for a Bruyères-le-Châtel data center with 13,800 GB300 chips and committed $1.4B to a 200MW Swedish facility.

2026-04-30

mistral-common v1.11 ships Mistral Guidance with lark grammar-guided generation

Mistral released mistral-common v1.11.0, introducing Mistral Guidance — lark-grammar-constrained generation for valid reasoning traces with or without tool calls. A v1.11.1 patch followed the same day to smooth agentic tool-message handling.

2026-04-29

Mistral Large 3 targets European on-prem and sovereign deployments

Mistral launched Large 3, a 200B-parameter model optimized for European enterprise with on-prem and sovereign-cloud deployments, positioning against GPT-5 on reasoning while emphasizing data residency. The release pairs with Mistral Small 4 (March), which unified reasoning, multimodal, and agentic-coding capabilities. Mistral also announced founding-member status in NVIDIA's Nemotron Coalition.

2026-04-27

DeepSeek V4 ratchets pressure: Mistral and Western open-weight labs face cost squeeze

DeepSeek V4's open-source release — Apache 2.0, 1.6T parameters, 1M context, $1.74/$3.48 per million tokens — surpasses Moonshot Kimi K 2.6 (1.1T) and MiniMax M1 (456B) on the open-weight leaderboard, intensifying pressure on Mistral, Llama 4, and Cohere to match price-performance.

2026-04-26

Mistral AI Releases Connectors in Studio with MCP Support and Forge Platform

Mistral AI launched Connectors in Studio, enabling developers to build enterprise AI applications grounded in proprietary data using built-in and custom Model Context Protocols. The company also unveiled Forge, a privacy-focused platform for building custom models trained exclusively on private data within company infrastructure, targeting $1 billion ARR by end of 2026.

2026-04-17

Mistral Medium 3 Released via API with 128K Context at Aggressive Pricing

Mistral AI launched Mistral Medium 3 through La Plateforme API, offering 128K token context at $0.40 per 1M input tokens. The model is positioned as a direct competitor to GPT-4o Mini and Claude 3.5 Haiku on price-performance for enterprise workloads. Mistral Medium 3 scores competitively on MMLU and MT-Bench while delivering faster time-to-first-token than its predecessor.

2026-04-16

Mistral Small 3.1 Adds Vision and 128k Context in Compact Open-Weight Release

Mistral AI released Mistral Small 3.1 (24B parameters) with native vision understanding and a 128k token context window under the Apache 2.0 license. The model is designed to run on a single GPU and outperforms GPT-4o Mini on several multilingual and vision benchmarks per Mistral's internal evaluations. It is available via Mistral's La Plateforme API and as a direct download on Hugging Face.

2026-04-14

Mistral AI Expands Product Line with Small 4, Voxtral TTS, and Forge Platform

Mistral AI rapidly expanded its product line in March 2026, with key releases including Mistral Small 4, a unified model encompassing reasoning, multimodal vision, and agentic coding capabilities. Other launches feature Voxtral TTS (open-weight text-to-speech) and Mistral Forge (an enterprise AI training platform). Mistral also joined the NVIDIA Nemotron Coalition as a founding member, deepening its partnership with NVIDIA.

2026-04-14

Mistral AI Releases Comprehensive Model Suite Including Small 4 and Voxtral TTS

Mistral AI unveiled multiple products between March 16 and March 31, 2026, including Mistral Small 4 which unifies Magistral, Pixtral, and Devstral capabilities into a single model. The company launched Voxtral TTS as an open-weight text-to-speech model, Leanstral formal proof agent, and Mistral Forge enterprise AI training platform announced at NVIDIA GTC on March 17. Mistral also joined the NVIDIA Nemotron Coalition as a founding member, collaborating on Nemotron 4 development and making Mistral Small 4 available on NVIDIA NIM and build.nvidia.com for broader deployment accessibility.

2026-04-13

Mistral Launches Small 4 MoE Model and Introduces Mistral Forge Enterprise Training Platform

Mistral AI released Mistral Small 4, a 119-parameter Mixture-of-Experts model with 128 experts and 4 active per token, benchmarked to match or surpass OpenAI's GPT-OSS 120B on long-context reasoning, live coding, and mathematics tasks. The company simultaneously launched Mistral Forge, an enterprise platform allowing organizations to train custom models on proprietary data, targeting the gap between generic foundation models and organization-specific deployments. This represents Mistral's push into the enterprise market while maintaining their focus on efficient model architectures.

2026-04-10

Mistral Releases Small 3.1 with Native Vision and Small 4 Hybrid Multimodal Model for Enterprise

Mistral released Mistral Small 3.1, a 24B parameter model adding native image understanding to the Small series for the first time, runnable on a single A100 GPU and available for self-hosting via Hugging Face or API via La Plateforme — positioned as a cost-efficient alternative for document OCR and visual QA tasks. Mistral also unveiled Mistral Small 4, a hybrid multimodal 119B parameter model (6B active per token) unifying reasoning, multimodal, and agentic coding capabilities with a reported 40% reduction in completion time versus its predecessor. Developers on X/Twitter noted Mistral Small 3.1 undercuts GPT-4o Mini on price for OCR tasks but raised questions about benchmark transparency.

2026-04-09

Mistral Small 4: 119B MoE Model with Reasoning, 256k Context, and Apache 2.0 License Matches GPT-OSS 120B

Mistral AI introduced Mistral Small 4, a Mixture of Experts model with 119B total parameters and a 256k context window, unifying capabilities from its Magistral (reasoning), Pixtral (multimodal), and Devstral (agentic coding) flagship models into a single release under Apache 2.0. The model features a configurable `reasoning_effort` parameter enabling developers to tune latency-quality tradeoffs, and matches or surpasses OpenAI's GPT-OSS 120B on long-context reasoning, live coding, and mathematics benchmarks while generating 20% less output on LiveCodeBench. Day-0 deployment on NVIDIA NIM microservices is accelerating enterprise adoption.

2026-04-08

Mistral CTO Guillaume Lample Teases Voxtral TTS and Mistral 4 Roadmap as Models Expand Across AWS Bedrock

Mistral co-founder and CTO Guillaume Lample discussed the upcoming Voxtral text-to-speech model, Forge, Leanstral, and plans for Mistral 4 in a recent podcast, signaling the European open-weight lab is diversifying beyond text-generation amid intense competition from Qwen and DeepSeek. Mistral models have been incorporated into AWS Bedrock's expanded catalog of nearly 100 models, and NVIDIA's GTC announcements included optimizations for Mistral Small 4, reinforcing Mistral's enterprise relevance. The company continues to distribute models under both Mistral Research License and Apache-2.0 options.

2026-04-06