The April 2023 founding
Mistral AI was incorporated in Paris on April 28, 2023 by three French AI researchers in their early thirties. Arthur Mensch (CEO) had been a researcher at Google DeepMind, where he co-authored the original LLaMA-related research; Guillaume Lample (Chief Science Officer) and Timothée Lacroix (CTO) had worked on the LLaMA paper at Meta. The three had attended École polytechnique together a decade earlier.
Mistral's seed round closed in June 2023 at €105 million ($117M USD) — an unusual scale for a four-week-old company, valuing the lab at roughly $260M pre-product. Lead investor was Lightspeed Venture Partners; the syndicate included Eric Schmidt, Xavier Niel (Iliad / Free), JCDecaux Holding, and Bpifrance (the French sovereign investment bank). The seed-round valuation drew sustained press coverage as a marker of European AI capital commitment.
The first model, Mistral 7B, shipped on September 27, 2023 — five months after incorporation. It was released via a magnet link posted to X, a deliberate stylistic choice that the lab repeated for Mixtral 8x7B in December 2023. The release-by-torrent pattern signaled an open-source identity that the subsequent licensing turn at the Microsoft partnership and the Mistral Research License would complicate.
The funding arc — from ASML's September 2025 Series C to Samsung's September 2026 Series D
After the June 2023 seed, Mistral raised a $415M Series A in December 2023 led by Andreessen Horowitz at a ~$2B valuation. A €600M (~$640M) round in June 2024 mixed equity and debt, led by General Catalyst, valued the lab at roughly $6B. The headline fundraise came on September 9, 2025: a €1.7 billion (~$2B) Series C led by Dutch semiconductor company ASML, with ASML contributing €1.3 billion for a ~11% fully-diluted stake alongside NVIDIA, Andreessen Horowitz, Bpifrance, DST Global, General Catalyst, Index Ventures, and Lightspeed, valuing Mistral at €11.7 billion (~$13.8B).
ASML's strategic position in EUV lithography for advanced semiconductor fabrication makes the investment substantively unusual: the deal is widely read as a Europe-on-Europe sovereignty bet, ASML buying a stake in the European frontier-AI tenant most likely to consume next-generation compute that ASML's lithography ultimately enables. The Series C held the record as Mistral's largest single equity round for just under a year, until the September 2026 Series D described below roughly doubled it. The lab has not publicly disclosed gross revenue or operating margin.
On March 30, 2026 Mistral closed an $830 million debt financing from a seven-bank consortium including Crédit Agricole CIB, HSBC, and MUFG (first investments in Mistral for each), earmarked for a 44 MW data center near Paris built on 13,800 NVIDIA GB300 GPUs, alongside Swedish capacity. The lab has publicly targeted 200 MW of European capacity by end-2027. See TechCrunch and CNBC.
On June 12, 2026, Bloomberg reported that Mistral was in early-stage talks to raise roughly €3 billion (~$3.5B) at a valuation of about €20 billion (~$23B) — nearly double the €11.7B Series C mark from September 2025 — reportedly to fund the lab's new “physics AI” push for industrial engineering (anchored by the May 2026 acquisition of physics-modeling startup Emmi). The talks firmed up through July: on July 22, 2026 the Financial Times reported that Samsung Electronics was in talks to invest up to €1 billion in the round, with Swedish investor EQT also reportedly participating, at the same roughly €20 billion valuation — specifically EQT's Scaleup Europe Fund, per the FT via Axios, which frames the talks against an industry-wide memory-and-semiconductor crunch. Follow-on reporting added Novo Holdings and Santander to the list of reported participants. The reporting was directionally right and low on both numbers: the round closed larger, at a higher mark, and with Samsung as lead rather than a €1 billion participant.
The round closed on September 8, 2026. Mistral announced a €3 billion Series D at a post-money valuation of more than €21 billion (~$24B) — which the lab describes as the largest equity fundraising round ever completed by a European technology company, and close to double the €11.7B Series C mark set a year earlier. Samsung Electronics led, joined by co-leads Scaleup Europe Fund (managed by EQT) and existing investor PSG Equity. Advent, funds and accounts managed by BlackRock, and the Grand Duchy of Luxembourg joined as new investors; existing backers a16z, ASML, Belfius, BNP Paribas CIB, Bpifrance, Carmignac, DST Global, Eurazeo, General Catalyst, Headline, Hillspire, Index Ventures, Korelya Capital, Lightspeed, NVIDIA, Phoenix Court's Solar fund, and Salesforce Ventures participated. Novo Holdings and Santander, both named in the mid-2026 follow-on reporting, do not appear in Mistral's published investor list. With ASML leading the Series C and Samsung the Series D, both of Mistral's last two rounds were led by an advanced-manufacturing corporate rather than a venture fund. Mistral says it now operates across 20 countries and supports 125+ enterprises, naming Airbus, ASML, and HSBC. Mensch told CNBC the proceeds go to infrastructure — building and owning data centers alongside rented capacity — with owned compute growing “around 100% in the next five years.” Sources: Mistral's announcement; Euronews.
A financing channel that is neither equity nor debt opened on August 11, 2026: Mistral began selling European Compute Units (ECUs) — multi-year customer commitments that convert into a claim on Mistral-built capacity, spendable on inference, training, or model adaptation as the buyer's needs change. Mistral says the aggregated demand is what determines which capacity gets built and where, and it named an anchor group including ASML, Amadeus, and Capgemini, whose CEOs are quoted in the announcement. The stated build target is the same 200 MW by end-2027 the lab has carried since the March 2026 debt raise, extended to up to 1 GW by 2030. No aggregate commitment value was disclosed.
The Microsoft partnership — February 2024 to the July 2026 compute expansion
On February 26, 2024, Microsoft and Mistral jointly announced a multi-year strategic partnership coinciding with the launch of Mistral Large on Azure AI. The deal carried a $16 million financial commitment from Microsoft and made Mistral models available natively on Azure as a preferred-partner offering, alongside OpenAI's models.
The partnership drew immediate French-government and EU regulatory scrutiny. The European Commission opened a brief antitrust look at whether the Microsoft commitment constituted a notifiable concentration under EU merger control; the Commission ultimately concluded it did not. Critics in Euronews coverage framed the deal as compromising the “European AI sovereignty” positioning the French government had used to justify its support of Mistral, including in the EU AI Act lobbying campaign covered in the next section.
Mensch's response (in a February 2024 Time interview and subsequent media appearances) framed the Microsoft deal as a commercial-distribution arrangement that did not affect Mistral's research direction or its open-source releases. The empirical record bears this out partially — the Mixtral 8x22B Apache 2.0 release in April 2024 followed two months later, and the Mistral 3 family relaunch in December 2025 re-committed the line to permissive open-source licensing across every scale — though the Mistral Large proprietary tier on Azure has remained a continuous commercial product.
On July 21, 2026, the partnership expanded substantially — and the infrastructure relationship now runs both ways. Microsoft announced a multibillion-dollar agreement to buy compute capacity from Mistral: Microsoft will lease part of Mistral's expanded Europe-based GPU infrastructure — built on thousands of NVIDIA Vera Rubin GPUs — to serve Azure cloud and AI demand, rather than only distributing Mistral's models on Azure. In the same announcement, Mistral Medium 3.5 and Mistral OCR 4 became available in Microsoft Foundry and Medium 3.5 in Microsoft Copilot Studio, deployable across cloud, cloud-connected, and fully disconnected Azure / Azure Local environments. Microsoft framed the deal under its Sovereign Cloud approach and its 2025 European Digital Commitments; no equity component was announced. Sources: Microsoft's July 21, 2026 release; France 24.
The EU AI Act lobbying narrative
Through late 2023 and into early 2024, Mistral mounted a sustained lobbying campaign against the European Parliament's tiered-foundation-model proposal in the EU AI Act, which would have imposed compliance and transparency obligations on general-purpose AI providers above defined capability thresholds. The campaign was led by Cédric O, France's former Secretary of State for Digital Affairs (2019–2022), who had joined Mistral as a strategic advisor and lobbyist after leaving government. Reporting at the time in Corporate Europe Observatory documented the privileged-access dynamic between Mistral and the highest levels of the French and German governments.
Mistral's stated position (in a November 2023 TechCrunch interview) was that the AI Act should regulate applications of AI rather than foundation models — that capability thresholds applied at the model-developer level would chill European open-source frontier-model development and entrench U.S. incumbents whose compliance teams were already large enough to absorb the regulatory cost. Critics countered that the “regulate applications, not models” framing aligned conveniently with the lobbying interests of the largest model developers themselves, which by 2025 included Mistral.
The final EU AI Act, agreed in December 2023 and entering force in August 2024, includes obligations on general-purpose AI providers but also includes broad exemptions for open-source models that do not pose “systemic risk.” The exemptions are widely credited — including by Corporate Europe Observatory's investigation — to the lobbying campaign Mistral led. The systemic-risk threshold, by contrast, captures the largest models and so applies to Mistral Large 2 / Mistral Large 3 onward; how the Code of Practice and the AI Office's compute-threshold methodology will treat each Mistral release is a continuous compliance question for the lab. Mistral said it would sign the GPAI Code of Practice within days of the Code's July 10, 2025 publication (EU Perspectives), and it did: Mistral AI appears on the Commission's published signatory list alongside Amazon, Anthropic, Google, IBM, Microsoft, OpenAI, Aleph Alpha, Cohere, and others — and as a signatory to the full code, not a chapter-limited one (xAI is listed as having signed only the Safety and Security chapter). Signing is voluntary, but the Commission treats adherence as the route to demonstrating compliance with the GPAI obligations.
The regime moved from obligation to enforcement on August 2, 2026. Per the Commission's own enforcement framework, the AI Office's and the national competent authorities' enforcement powers apply from that date: the AI Office may request information from providers, require access to a general-purpose model for evaluations run by its own staff or appointed independent experts, order a provider to address identified risks, and where necessary restrict a model's public availability. Breaches of the general-purpose-AI obligations carry fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher — against a ceiling of €35 million or 7% for the prohibited-practices tier. The AI Office has said its preferred first instrument remains a technical compliance dialogue rather than a formal proceeding. The same date switched on a second, separate obligation that reaches Mistral's consumer surface rather than its weights: the Article 50 transparency rules now require interactive AI systems to disclose that a user is talking to a machine, deepfakes to be labelled, and AI-generated or altered content to carry machine-readable marks. The Commission published a companion Code of Practice on Transparency of AI-generated Content on July 31, 2026, which about 190 organisations had signed by the end of that month. It is a distinct instrument from the GPAI Code, and Mistral signed it too: the Commission's signatory list names Mistral both among its Section 1 examples (alongside Aleph Alpha, Anthropic, Black Forest Labs, Cohere, Google, Meta, Microsoft, OpenAI, Synthesia) and in the full roster. That list lives on a separate news page rather than on the code's own policy page, which is why an earlier version of this paragraph said Mistral's adherence was not confirmable from the primary source; it is, and the claim was wrong from the moment the roster was published. Corrected September 24, 2026. As of October 30, 2026 no AI Office action naming Mistral has been made public. The Commission's enforcement-framework page now carries an October 20, 2026 stamp, and its “Monitoring tools” section now lists three channels: the AI Act Complaints Tool, an AI Act Whistleblower Tool, and a complaints channel for downstream providers that build on another provider's GPAI model. It still describes the request-for-information power only in the abstract. The AI Office's own page (re-stamped October 29, 2026) names no provider either, and the newsroom AI items it lists for the window, such as an October 5 special meeting of the Scientific Panel on frontier AI safety and risks and late-October cybersecurity-and-AI events, name none; the lab's systemic-risk-tier releases from Mistral Large 2 onward are the ones the general-purpose-AI half of the regime reaches. Read that stamp narrowly: requests for information are confidential supervisory correspondence, so the Commission's public surfaces return the same bytes whether or not a provider has been contacted — an absence of published action is not evidence of an absence of action.
The licensing turn — from MRL back to Apache 2.0
Mistral's licensing has evolved across three distinct conventions. The founding lineage — Mistral 7B (September 2023), Mixtral 8x7B (December 2023), Mixtral 8x22B (April 2024), Mistral NeMo (July 2024), Codestral Mamba and Mathstral (July 2024), Pixtral 12B (September 2024) — shipped under the standard Apache 2.0: weights public, commercial use unrestricted. In parallel, the proprietary tier (Mistral Medium December 2023, Mistral Large February 2024 onward) shipped closed-weights via la Plateforme.
The licensing innovation arrived in May 2024 with Codestral 22B, which introduced the bespoke Mistral AI Non-Production License (MNPL): weights public on HuggingFace, but commercial / production deployment requires a separate Mistral Commercial License. The MNPL pattern was generalized into the Mistral Research License (MRL), which was applied to Mistral Large 2 (July 2024, and its November 2024 update Mistral Large 2.1), Mistral Small 2 (September 2024),Ministral 8B (October 2024 — the edge model whose weights Mistral released “for research use,” while the sibling Ministral 3B stayed closed), and Pixtral Large (November 2024). The MRL/MNPL pattern is the one tracked by the violet “Research” family pill on this page.
The licensing turn arrived on January 30, 2025 with Mistral Small 3, which returned to Apache 2.0. Most subsequent Mistral open releases have shipped under Apache 2.0: Mistral Small 3.1, Mistral Small 3.2, Devstral, Devstral Small 1.1, Devstral Small 2, Magistral Small, Mistral Large 3, the Ministral 3 family (3B / 8B / 14B), Mistral Small 4, the Apache half of the Voxtral audio line (Voxtral Small, Voxtral Mini, Voxtral Mini Transcribe Realtime), Leanstral / Leanstral 1.5, and Shieldstral 1.0 (August 2026). The December 2025 “Mistral 3” family relaunch formalized the Apache 2.0 commitment across every scale — including, notably, the 675B-total-parameter Mistral Large 3 frontier flagship, which is still the largest Apache 2.0 Mistral model and was explicitly framed at launch as a return to permissive open-source. It was also the largest downloadable Mistral model until Mistral Large 4's weights shipped on October 29, 2026 under a different license, described below.
A growing set of releases sits outside the Apache 2.0 default without being proprietary, which is the reason the Family column's three pills no longer map one-to-one onto license names. Devstral 2 (December 2025), Mistral Medium 3.5 (April 2026), and the Timestral time-series family (October 2026) ship open weights under a Modified MIT license — the docs model cards carry a “Modified MIT” badge and the HuggingFace repos are tagged license:other rather than apache-2.0. The modification matters: the LICENSE file in each repo grants the standard MIT rights and then withholds all of them from any company (or employer) whose global consolidated revenue exceeded $20 million in the preceding month, derivatives included, unless Mistral grants a separate commercial license — so it is permissive for individuals, startups, and research, and closer to the MRL for large enterprises. Voxtral TTS (March 2026) ships open weights under CC BY-NC 4.0, which bars commercial use outright. Mistral Large 4 brought a third variant. It launched on October 6, 2026 as a preview API whose card said only “Open,” and when its weights shipped on October 29 they came under a new Mistral Ecosystem Open-Weight License. That license is also a modified MIT, but it has no revenue cap. Instead, any licensee running a “Model as a Service” business, meaning one that gives third parties hosted inference or post-training access to the model, needs a commercial license from Mistral. Internal use and end-user products that embed the model are exempt. This page files the Modified MIT releases and Large 4 under the sky Open pill (commercial use is permitted, with each license's carve-out noted on its row) and Voxtral TTS under the violet Research pill (weights public, commercial use restricted); read the per-model license text before shipping at scale rather than inferring it from the pill.
The proprietary tier has run in parallel throughout: the original Mistral Small (February 2024, alongside Mistral Large), Mistral Saba (February 2025), Codestral 25.01 (January 2025) and Codestral 25.08 (July 2025), Mistral Medium 3 (May 2025) and Mistral Medium 3.1 (August 2025), Magistral Medium 1.0–1.2 (June–September 2025), the whole OCR line (OCR, OCR 2, OCR 3, OCR 4, OCR 4.1), Voxtral Mini Transcribe 2 (February 2026), Mistral Moderation 2 (March 2026), and the Voxtral 3 audio family (October 2026) all shipped closed-weights via la Plateforme — Mistral's docs badge these “Premier.” Most of that roster has since retired; the Premier endpoints still served as of October 30, 2026 are Codestral 25.08, OCR 3 and OCR 4.1 (OCR 4.0 retired September 30, 2026, outlived by the older OCR 3), Voxtral Mini Transcribe 2, the four Voxtral 3 models (October 2026, public preview), Mistral Moderation 2, and the two embedding models (Mistral Embed, Codestral Embed), which get no row here. Voxtral TTS is not one of them: it is API-priced but its weights are published, under the non-commercial CC BY-NC 4.0 terms described above, which is why it carries the violet Research pill rather than the amber Proprietary one. The hybrid model — Apache 2.0 open releases for the public-facing flagship lineage, proprietary closed-weights for specialized commercial products (including the entire Codestral line from version 25.01 onward) — is the licensing equilibrium Mistral has settled into.
The 2026 release cadence and the NVIDIA partnership
Mistral's release rhythm accelerated through early 2026. In March 2026 alone, the lab shipped six major products in roughly two weeks: Mistral Small 4 (March 16), the Devstral Vibe CLI, Voxtral TTS (March 23, the lab's first audio model), Forge, the Spaces CLI, and a publicly-announced NVIDIA partnership for inference-stack optimization across the Mistral lineage on NVIDIA Blackwell hardware.
The cadence reflects the engineering capacity unlocked by the September 2025 ASML round and the Microsoft / NVIDIA / hyperscaler distribution deals collectively. Whether the cadence is sustainable through 2026 is an open question; the prior Mistral release rhythm averaged roughly 6–10 model rows per year, and the March 2026 wave represented an outlier rather than a new steady state. After Mistral Medium 3.5 (April 28, 2026), the spring focus shifted from model launches to product surface, with the changelog recording platform work (Vibe, Connectors, Workflows, Search Toolkit). The next model release was Mistral OCR 4 (June 23, 2026) — a specialized document-intelligence model, not a general-purpose LLM — followed by a Mistral OCR 4.1 point release on July 16, 2026 that shipped with no blog post at all, documented by a docs model card, a one-line changelog entry, and a legal-center family page the card does not link to. On July 8, 2026 the lab shipped Robostral Navigate (mistral.ai/news/robostral-navigate), its first embodied-navigation model and the first shipped product of the physics-AI push begun with the May 2026 Emmi acquisition — an 8B vision-language model that steers a robot from a single RGB camera and a plain-language instruction, again a specialized model rather than a general-purpose LLM. The NVIDIA relationship deepened past inference-stack optimization on July 21, 2026, when the expanded Microsoft agreement (covered in the Microsoft section above) had Mistral deploying thousands of NVIDIA Vera Rubin GPUs in its European data centers, part of whose capacity Microsoft will lease for Azure. It widened again the following week: on July 27, 2026 NVIDIA launched the Open Secure AI Alliance, a group organized around open models, harnesses, and tools for AI security, and Mistral is named among the inaugural partners alongside Microsoft, IBM, Hugging Face, Red Hat, Intel, Samsung Electronics, and the Linux Foundation. Mistral's first contribution followed on August 4, 2026 — Shieldstral, a 3B Apache 2.0 multimodal safety classifier and the lab's first open-weights guardrail model.
The strategic emphasis through August 2026 moved from models to infrastructure and retrieval. On August 11, 2026 Mistral announced a three-part sovereign-AI package: Mistral Regional Endpoints went generally available (customers choose whether inference runs in Europe or the US); a Mistral Priority Tier with custom rate limits and an uptime SLA entered public preview; and the lab opened an anchor coalition of enterprises whose multi-year European Compute Unit (ECU) commitments are meant to underwrite up to 1 GW of European capacity by 2030. The same post carried a first for the platform: Mistral will serve third-party open models alongside its own, starting with Z.ai's GLM 5.2 (zai-glm-5-2, a 1M-context text model dated August 6, 2026 on its docs card, served without Mistral modifications). That line has since grown and turned over: a GLM 5.3 card (zai-glm-5-3) appeared dated September 15, 2026 on the same Third-party / Open badges, went generally available on September 28, and the next day GLM 5.2 was deprecated with an October 31, 2026 retirement date — so the third-party line already has a Legacy entry of its own in the docs' Deprecated & retired table. The set kept growing in October: Z.ai GLM 5.3 Flash (zai-glm-5-3-flash, a 320B-total / 18B-active multimodal MoE, changelog and card dated October 22, 2026, GA) and DeepSeek V4.1 Flash 1017 (deepseek-v4-1-flash-1017, card dated October 28, 2026, Public Preview, with no changelog entry as of October 30). That made DeepSeek the second third-party lab on la Plateforme. None of the third-party models is a Mistral model, so none gets a row on this page, but GLM 5.2 was the first non-Mistral model la Plateforme hosted and the docs card's Third-party badge is the gate that keeps them out of this table. Nine days later, on August 20, 2026, Mistral shipped Agentic Search — a multi-step retrieval loop built into the Mistral Search Toolkit and Libraries that gives a model five file-system-shaped tools (search, open, navigate, read, grep) over an existing index, reporting FinanceBench correctness rising from 26.7% to 86%. Both are platform releases rather than model releases, so neither adds a table row. The month's only model-side event was a lifecycle change rather than a launch: on August 31, 2026 a one-line changelog entry moved Mistral OCR 4.1 from Public Preview to general availability, six and a half weeks after it shipped and still with no blog post of its own.
The sovereignty push then went extra-European. On August 24, 2026 Mistral announced a strategic collaboration with HUMAIN — the full-stack AI company owned by Saudi Arabia's Public Investment Fund — covering AI infrastructure, model development, and deployment in Saudi Arabia and across the region, which the two companies describe as “a collaboration in the hundreds of millions of Euros” (mistral.ai/news/mistral-x-humain). The stated initial focus is cybersecurity and voice, with a plan to develop frontier models that perform strongly in Arabic; Mistral says it will explore running on HUMAIN's data-center infrastructure to serve local compute demand, and the two will build a joint go-to-market for regulated industries. Mistral's own framing puts it in a sequence with the July 2026 Microsoft compute expansion and the August 11 European Compute Units launch — the same sovereign-AI argument exported rather than a new one. Nothing has shipped from it yet: no model, no API string, no HuggingFace repo, so no table row. If the Arabic frontier-model work does ship, it would be the first successor to Mistral Saba, the regional model retired in September 2025 with no replacement.
The same argument reached the enterprise data stack two weeks later. On September 10, 2026 Mistral and Cloudera announced a partnership (mistral.ai/news/mistral-x-cloudera) with two stated parts: Mistral's models get integrated into Cloudera's hybrid data platform so enterprises can run inference across private cloud, on-prem, and fully air-gapped environments; and customers can train adapted models against proprietary data inside those same controlled environments, keeping both the data and the resulting weights. Mistral's Kamal Brar frames the reach as Cloudera's 30 exabytes of customer-managed data. No financial terms, no model, no API string, and no HuggingFace repo, so no table row — but it extends the same open-weights-plus-customer-controlled-infrastructure pitch that runs through the July 2026 Microsoft compute expansion, the August 11 European Compute Units launch, and the August 24 HUMAIN collaboration.
Six days later the distribution argument reached consumers for the first time. On September 16, 2026 Mistral and Mozilla announced a partnership (mistral.ai/news/mistral-x-mozilla) putting Mistral models behind Firefox Smart Window, Mozilla's AI browsing assistant, in beta. Mistral powers Smart Window for users in France and North America at launch, with the United Kingdom and Germany said to follow later in 2026, and the two describe fine-tuning on regional languages, dialects, and cultural context rather than exporting a single model everywhere. The privacy terms are the part with teeth: Smart Window conversations are not saved on Mozilla's servers by default, and Mistral is named as agreeing to zero data retention. Mistral frames it as its first move past the enterprise into consumer reach — every prior 2026 partnership on this page (Microsoft, HUMAIN, Cloudera) sold compute or enterprise deployment. No model, no API string, and no new HuggingFace repo came with it, so no table row; the announcement names no specific Mistral model behind Smart Window.
The physics-AI push got an address twelve days later. On September 28, 2026 Mistral opened a hub in Munich (mistral.ai/news/hallo-deutschland) housing research teams for Physics AI and Industrial AI alongside applied engineers who work with enterprise customers directly, and credits the 30-plus physicists, researchers, and engineers who arrived with the May 2026 Emmi acquisition. The post names two German engagements — BMW on crash simulation and engineering AI, and Siemens Energy on industrial AI applications — plus a research partnership with the Technical University of Munich that uses TUM's wind tunnel to build digital twins for automotive aerodynamics. It restates the one-gigawatt-of-European-compute-by-2030 target from the August 11 European Compute Units launch and quotes Germany's Federal Minister for Digital Transformation, Karsten Wildberger. An office and a set of customer engagements rather than a release: no model, no API string, and no HuggingFace repo, so no table row.
Then, after five months of platform work, partnerships, and specialized models, the flagship line moved. On October 6, 2026 Mistral put Mistral Large 4 into public preview (mistral.ai/news/mistral-large-4) — a 52B-active / 1.05T-total granular MoE with a 1M context, trained on 3,800 NVIDIA Grace Blackwell GPUs in the lab's own European data centers and framed by Mistral as the first milestone of the Series D roadmap. It launched API-only, with the weights promised “by the end of the month,” and Mistral kept that promise: on October 29, 2026 it released the weights of an improved checkpoint after three more weeks of reinforcement learning (mistral.ai/news/mistral-large-4-oss), under a new Mistral Ecosystem Open-Weight License that restricts Model-as-a-Service hosting, with Microsoft Foundry named as the first cloud to carry it. By then the API list price had fallen to $0.70 / $2.10 per million tokens, about half the launch list price. Nine days after the preview, on October 15, 2026, came Timestral (mistral.ai/news/timestral), the lab's first time-series foundation models — four sizes from 5M to 209M parameters, Modified MIT, CPU-only inference, the first Mistral release that is not a language model at all and the clearest product of the physics-and-industrial-AI push that the Emmi acquisition and the Munich hub set up. Between the two, the HuggingFace org picked up weights-only Arabic derivatives of the Voxtral Realtime model on October 8 (folded into the Voxtral V2 row), and a September 29, 2026 batch of changelog deprecations retired Leanstral 1.5, OCR 4.0, and the third-party GLM 5.2 endpoints on short notice. On October 20 came Voxtral 3 (mistral.ai/news/voxtral-3), four closed-weights speech-to-text and text-to-speech models plus a managed Voice Agents pipeline. It is the first Voxtral generation without public weights. The next day, October 21, Mistral and NVIDIA announced a collaboration on AI-native engineering workflows (mistral.ai/news/mistral-nvidia-ai-for-industrial-engineering) that pairs Mistral Large 4 and Mistral's physics-modeling pipelines with NVIDIA CUDA-X, Warp and Omniverse, and promises public benchmarks and open datasets for physics and agentic AI in computer-aided engineering. That is a partnership, not a release, so it adds no row. The cadence question the March wave raised looks settled from here: Mistral ships a general-purpose flagship roughly twice a year and fills the months between with specialized models and platform surface.
On May 28, 2026, at its inaugural AI Now Summit in Paris, Mistral rebranded Le Chat as Vibe (mistral.ai/news/vibe-agent) — a single agent spanning Work mode (long-horizon productivity across Google Workspace, Outlook, SharePoint, Slack, and GitHub connectors), Code mode (remote coding agents and a VS Code extension), and a legacy Chat mode that preserves the prior Le Chat experience and history. The summit also introduced an industrial “physics AI” stack — built on the May 2026 acquisition of Emmi — targeting aerospace, automotive, and semiconductor engineering, with Airbus, BMW, and ASML named as customers. Both are product / platform developments rather than language-model releases, so neither adds a table row.
Where to run Mistral
Mistral is widely deployed because most flagship releases are open-weights and the proprietary releases are available across every major hyperscaler. Inference paths through 2025–2026 break into four categories.
Mistral's own surfaces. Vibe — the rebrand of Le Chat announced May 28, 2026 (mistral.ai/news/vibe-agent) — is the consumer + agent product (free / Pro $14.99 / Team $24.99 / Enterprise, iOS / Android / web), with Work, Code, and Chat modes; existing Le Chat history, settings, and plans carried over. la Plateforme (Mistral Studio) is the developer API endpoint (OpenAI-compatible).
Self-host from HuggingFace. Download from the mistralai org and run with vLLM, llama.cpp, Ollama, or NVIDIA TensorRT-LLM. Check the license before you deploy, because the open-weights tier is no longer uniformly Apache 2.0: the Apache releases (Mistral 7B through Mistral Small 4, plus Leanstral 1.5 and Shieldstral 1.0) self-host without commercial restriction; Mistral Medium 3.5, Devstral 2, and Timestral ship under Modified MIT, which permits commercial use only for companies under $20 million in monthly revenue (above that, a separate Mistral commercial license); Mistral Large 4 (weights since October 29, 2026) ships under the Mistral Ecosystem Open-Weight License, which allows commercial self-hosting but requires a separate commercial license before you offer it to third parties as a hosted model service; the MRL / MNPL releases (Codestral 22B, Mistral Large 2, Mistral Small 2, Ministral 8B, Pixtral Large) require a separate Mistral Commercial License for production use; and Voxtral TTS is CC BY-NC 4.0, so the downloadable weights are research-only and commercial voice generation has to run through la Plateforme.
Hyperscalers. Microsoft Azure AI Foundry (the Microsoft launch partner since February 2024), AWS Bedrock, Google Cloud Vertex AI, IBM watsonx, Oracle OCI, NVIDIA NIM (the Mistral NeMo launch partner since July 2024). Most hyperscalers carry both the proprietary and the open-weights lineage; the per-model availability matrix lives at docs.mistral.ai.
Hosted-inference providers. Together AI, Fireworks, OpenRouter, Replicate, Groq. Most providers serve the Apache 2.0 lineage with similar latency / cost characteristics; the MRL-licensed weights (Mistral Large 2, Pixtral Large) typically require an additional commercial-license attestation from the provider.
People who shaped Mistral
Arthur Mensch — co-founder and CEO. PhD from École normale supérieure / INRIA; researcher at Google DeepMind 2020–2023 working on RETRO and Chinchilla. The face of Mistral in EU and US policy debates; profiled in Time and the 20VC podcast.
Guillaume Lample — co-founder and Chief Science Officer, the title on Mistral's own About page. PhD from Sorbonne / Facebook AI Research; co-author of the Meta LLaMA paper. Leads the model-architecture and pretraining work behind every Mistral release; first author on the Mistral 7B paper.
Timothée Lacroix — co-founder and CTO. PhD from École polytechnique / Facebook AI Research; co-author of the Meta LLaMA paper. Leads the engineering and infrastructure work, including the la Plateforme API, the inference stack, and the hyperscaler integrations.
Cédric O — strategic advisor and lobbyist. France's former Secretary of State for Digital Affairs (2019–2022); has led Mistral's EU AI Act lobbying campaign since 2023, with documented privileged access to the highest levels of French and German government per Corporate Europe Observatory's reporting. Not a corporate officer; Mistral's regulatory voice in Brussels.
The competitive landscape
Mistral occupies a distinctive position: the leading European frontier-AI lab, with a hybrid commercial strategy that puts open-weights flagships (Apache 2.0 or Modified MIT from Mistral Small 3 onward) alongside a proprietary API tier on la Plateforme and the hyperscalers. The closest open-weights competitors are Meta's Llama (custom Llama Community License with the >700M-MAU carve-out, see Llama Versions), DeepSeek (Chinese, MIT-licensed for the V3 / R1 line and onward, see DeepSeek Versions), and Alibaba's Qwen (Apache-2.0-or-permissive across most releases — see Qwen Versions). The closed-weights frontier competitors — ChatGPT, Claude, Gemini, Grok — have all stayed closed-weights since their inception. Mistral's regulatory and political position in Europe is the variable that distinguishes the line from every other frontier lab; whether the EU AI Act's Code of Practice and the AI Office's enforcement methodology will treat European model providers more or less stringently than U.S. and Chinese ones remains the open structural question for the lab through 2026. This page does not attempt a benchmark roundup or a ranking.