Signals Inbox·July 28, 2026·Frontier AI

How big is Mistral vs. OpenAI or Anthropic?

Mistral has become Europe’s most credible AI challenger, but financially it is still only about 1% to 2% the size of OpenAI or Anthropic, with its real opening concentrated in private, customizable enterprise AI.

Q1So how big is Mistral next to OpenAI and Anthropic?

Mistral is currently about 1% to 2% as large as OpenAI or Anthropic on the financial measures that matter most.

Mistral’s latest completed funding round valued the company at €11.7 billion, equivalent to roughly $13.5 billion when the deal was announced. OpenAI’s latest round valued it at $852 billion, while Anthropic has since reached $965 billion. Mistral therefore represents around 1.4% to 1.6% of either company’s valuation.

Revenue tells an even harsher story. Mistral’s annual recurring revenue has recently been estimated at around $400 million. OpenAI says it now generates about $2 billion every month, equivalent to a $24 billion annual pace. Anthropic’s latest official figure is $47 billion in annualized revenue.

Mistral has recently entered discussions over another round that could value it at roughly €20 billion. Even if that deal happens, Mistral would remain worth only around 2% of its two US rivals.

So the simple answer is clear. Mistral has become a major European technology company within three years, but it is still a small challenger beside two companies approaching trillion-dollar valuations.

Mistral, OpenAI and Anthropic compared

Q2How much money is Mistral actually making today?

Mistral is making serious money for a three-year-old European startup, but its current revenue still belongs in a different league.

Its estimated $400 million annual run rate would make Mistral a substantial enterprise software company. The figure also appears to have risen extremely quickly from a much smaller base. OpenAI now generates roughly the same amount in six days. Anthropic gets there in about three.

That is the practical gap: the two US leaders can reinvest more, hire faster, subsidize products and reserve far more computing capacity.

Mistral is reportedly aiming to pass $1 billion in annual recurring revenue by the end of the year. Reaching that target would confirm that its large industrial contracts are turning into a repeatable business. It would also make Mistral one of Europe’s fastest-growing software companies.

Even at $1 billion, Mistral would generate only about 4% of OpenAI’s current revenue pace and 2% of Anthropic’s. It can grow spectacularly without closing much of the gap.

Q3Is Mistral catching up, or are OpenAI and Anthropic pulling away?

Mistral is growing extremely fast, yet the absolute gap is still widening.

Estimates suggest that Mistral’s annual recurring revenue rose roughly 25-fold from the end of 2024 to its current level. Few software companies ever grow that quickly.

Anthropic began from a much larger base and accelerated even harder. Its annualized revenue increased from around $9 billion at the end of 2025 to $47 billion within months. It added roughly $38 billion of annual revenue pace during the period, close to 100 times Mistral’s entire current business.

OpenAI has followed the same broad pattern. It was producing about $1 billion per quarter at the end of 2024. It now produces around $2 billion per month, giving it roughly six times the annualized revenue it had less than two years ago.

Percentage growth makes Mistral look closer than it is. The dollars added show OpenAI and Anthropic building their leads faster than Mistral is reducing them.

Fast growth by itself will not do it. Mistral would need a market shift: self-hosted AI becoming the default for large companies, governments moving heavily toward European suppliers, or one of its open models becoming a global standard.

Q4Can Mistral raise enough money to keep playing?

Mistral can raise billions, but OpenAI and Anthropic can now raise tens of billions in a single round.

Mistral’s €1.7 billion Series C was one of the largest technology financings ever completed in Europe. ASML led the round and became its largest shareholder, giving Mistral both capital and a close relationship with one of the world’s most important semiconductor companies.

OpenAI later raised $122 billion. Anthropic raised $65 billion. At rough exchange rates, those rounds were about 60 and 30 times larger than Mistral’s.

Mistral’s latest fundraising discussions suggest that investors remain interested. Samsung is reportedly considering an investment as part of a broader round at a valuation near €20 billion. That deal has not closed, so the valuation is not established.

Even several additional billions would leave Mistral with much less room for mistakes. OpenAI and Anthropic can fund several training programs, infrastructure projects and product launches at once. Mistral has to choose.

Its investors point toward the strategy. ASML can become a customer and technical partner. Samsung could provide another route into chips, devices and Asian markets. Mistral needs industrial advantages alongside the money.

Q5Does Mistral have enough computing power to compete?

Mistral has enough computing power to build competitive models, but nowhere near the reserve capacity available to OpenAI and Anthropic.

Mistral trained Large 3 from scratch using 3,000 Nvidia H200 GPUs. That is a serious training cluster, especially for a company founded in 2023, and it helped produce one of the stronger permissively licensed open models.

The US leaders now operate on a much larger infrastructure base. OpenAI says its announced Stargate sites represent more than eight gigawatts of planned capacity and over $450 billion of investment. Separate agreements cover additional Nvidia, AMD and custom Broadcom systems, although these projects overlap and should not simply be added together.

Anthropic currently uses more than one million Amazon Trainium2 chips. Its expanded Amazon agreement covers over $100 billion of spending and up to five gigawatts of additional capacity. It is also securing large amounts of Google TPU capacity.

A single training run and a multiyear data-center program are different measurements. Still, the scale gap is hard to miss. OpenAI and Anthropic can run more experiments, train larger models and serve much heavier demand.

Mistral has to compete through efficiency: cheaper training, easier deployment and models that are strong enough for specific jobs without always chasing the largest possible system.

Q6Are Mistral’s models close to GPT and Claude today?

Mistral’s best current model can handle real enterprise work, but the frontier still belongs to OpenAI and Anthropic.

Artificial Analysis currently gives Mistral Medium 3.5 a score of 30 on its combined Intelligence Index. OpenAI’s strongest GPT-5.6 configuration scores 59, while Claude Opus 5 reaches 61.

The index combines difficult tests involving coding, science, professional work, reasoning and autonomous agents. A gap close to 30 points is substantial. GPT and Claude are generally more reliable when a task becomes vague, long, unfamiliar or dependent on several correct decisions.

Mistral Medium 3.5 still clears the bar for many everyday jobs. It handles text, images, coding, reasoning and tool use. It supports a 256,000-token context window and can run on as few as four powerful GPUs.

That makes it practical for document extraction, internal search, classification, customer support, routine coding and tightly defined agents. A company processing invoices or searching engineering documents may gain little from paying for the smartest available model.

The gap becomes harder to ignore in open-ended research, complicated software engineering or long autonomous projects. There, the stronger model may simply finish more often.

Current model capability and pricing comparison

Q7Is Mistral cheap enough to win anyway?

Price gives Mistral a real opening, although cheap tokens cannot rescue a model that fails the task.

Mistral Medium 3.5 costs about 70% less than the strongest GPT and Claude models at published API prices. For a business processing millions of repetitive requests, that difference can save meaningful money.

The discount shrinks when we measure the cost of completing a task rather than producing a token. Artificial Analysis recently estimated an average benchmark-task cost of about $0.56 for Mistral Medium 3.5, compared with $1.04 for GPT-5.6 Sol and $2.03 for Claude Opus 5 at their highest settings.

Mistral remained cheaper, but the advantage was closer to two or four times rather than the raw token-price gap. Weaker models may need longer prompts, more retries or extra verification. Sometimes the cheap model is not the cheap option.

The likely setup for many companies is a mix. Mistral can handle high-volume, predictable and sensitive work. GPT or Claude can take over when the request becomes unusually difficult.

Mistral does not need to replace every OpenAI or Anthropic request. Capturing the large layer of routine enterprise work could support a very large business.

Q8Does Mistral have real customers, or mainly famous partnerships?

Mistral has real production customers, although its public evidence still points to a concentrated enterprise business.

ASML is using Mistral for software reliability, large-scale log analysis and technical documentation. CMA CGM has deployed it across shipping, logistics and media tasks, including cargo release, customer requests and document processing.

Airbus is working with Mistral across commercial aviation, helicopters, defense and space, with some systems designed for on-premise deployment. SNCF says its internal GPT system reaches more than 100,000 employees, while its developers also use Mistral’s coding tools.

These are serious environments, not lightweight chatbot trials. Semiconductor manufacturing, aerospace engineering, banking and transport all impose hard requirements around security, accuracy and data access.

What we do not see is a customer count comparable with the US leaders. OpenAI says more than one million business customers use its products. Anthropic has reported more than 300,000 business customers, with over 1,000 accounts now spending at least $1 million annually.

Mistral highlights a smaller collection of large deployments but publishes no similar total. Its strength seems to come from deeper relationships with selected companies, especially in Europe, rather than broad distribution across hundreds of thousands of businesses.

Q9Is Mistral’s Vibe anywhere close to ChatGPT or Claude?

Vibe is far smaller than ChatGPT and Claude, and Mistral now treats it mainly as a work-and-code agent rather than a mass-market chatbot.

Mistral recently renamed Le Chat as Vibe. The product combines chat, research, workplace actions and coding under one subscription. It can run long tasks, connect to business tools and launch remote coding agents.

The shift says a lot about Mistral’s priorities. It is aiming at people who want an agent to complete work, especially developers and enterprise teams. Casual consumer chat is no longer the center of the pitch.

OpenAI already has more than 900 million weekly ChatGPT users and over 50 million paying consumer subscribers. Codex alone has around three million weekly users.

Anthropic also has a large professional audience. Claude Code has passed $2.5 billion in annualized revenue, and its weekly user count doubled during the first part of the year.

Mistral does not publish a current active-user figure for Vibe that can be compared with those numbers. Earlier download milestones showed interest, but downloads say little about regular use or paid retention.

Vibe can still help Mistral sell models and enterprise deployments. It gives users a simple way to experience the technology and creates a direct channel into coding and workplace automation. Consumer scale, though, remains overwhelmingly in OpenAI’s favor.

Q10Why would a company choose Mistral over a stronger US model?

A company picks Mistral when control, deployment and customization matter more than having the strongest model on every task.

Mistral lets customers run models through its cloud, another cloud provider, a private cloud or their own servers. Some open models can also be downloaded, modified and operated without sending every request back to Mistral.

That flexibility appeals to banks, defense organizations, factories and public institutions. They may have sensitive data, old internal systems or rules that prevent information from leaving a controlled environment.

Customization can also go deeper. A company can fine-tune a Mistral model on its terminology, connect it tightly to internal tools and optimize it around one workflow. A specialized model may outperform a larger general model on the exact job it was built to handle.

OpenAI and Anthropic provide strong privacy controls, regional processing and enterprise contracts. They can satisfy many regulated customers without offering downloadable versions of their strongest models.

Mistral’s advantage is clearest when the customer wants direct control over the model itself. That market is smaller than the full AI market, but its buyers can sign large, sticky contracts.

Q11Can Mistral make money when customers can download its models?

Mistral can make money from open models, but it has to sell much more than access to model weights.

A customer that downloads a Mistral model may avoid paying the company for every token. Cloud providers and software developers can also build services around the model, spreading Mistral’s technology without automatically increasing its revenue.

Mistral is building several paid layers around that open base. It sells hosted API access, private deployments, model customization, technical support, enterprise software and computing capacity through Mistral Compute.

The open models can work as distribution. Developers test them for free, companies build prototypes, and some later pay Mistral to deploy the system securely or operate it at scale.

The risk is obvious. Meta, OpenAI, Chinese laboratories and independent developers also release open-weight models. If comparable models remain widely available, customers may switch easily or ask another provider to host them.

Mistral must prove that its engineering support, enterprise products and infrastructure are worth paying for. Popular downloads alone will not finance a company valued in the tens of billions.

Q12Can Europe make Mistral a global winner?

Europe can give Mistral a large home base, but Europe alone cannot make it a peer of OpenAI or Anthropic.

The European Union has more than 450 million residents, large governments and world-leading companies in banking, aerospace, automotive manufacturing, energy, pharmaceuticals and industrial equipment.

Many of those organizations care about data location, regulatory control and dependence on foreign suppliers. Mistral fits those concerns better than almost any other European AI company.

Its strongest customer relationships already follow this pattern. ASML, Airbus, BNP Paribas, CMA CGM, SNCF and Stellantis operate in industries where security and technological control can influence purchasing decisions.

Still, European companies regularly buy American technology when it works better or integrates more easily. Microsoft, AWS, Google and Nvidia already sit deep inside their infrastructure. OpenAI and Anthropic reach those customers through platforms they already use.

Political support may help Mistral enter negotiations, win government work or secure financing. It cannot compensate indefinitely for weaker models, difficult deployment or poor economics.

Mistral needs customers outside Europe too. Its recent discussions with Samsung make sense in that context. A global Mistral would need major Asian, Middle Eastern and American clients alongside its European industrial base.

Q13What would have to change for Mistral to become a real peer?

Mistral becomes a real peer only if it owns one large category worldwide.

Trying to copy every part of OpenAI would spread Mistral too thin. ChatGPT already dominates consumer distribution, while Anthropic has built exceptional momentum in coding and professional AI.

The clearest opening lies in private and self-hosted enterprise AI. Mistral could become the default provider for organizations that want capable models without handing control of their systems to a US platform.

That would require several billion dollars of recurring revenue, repeated expansion inside existing customers and a much broader international client base. Mistral would also need to remain close enough to the frontier that customers do not feel they are sacrificing too much capability.

Its computing business would have to grow as well. A company promising technological independence cannot rely entirely on rivals and foreign cloud providers to train and run its products.

Then watch what customers do after the first deployment. One successful document-search tool proves Mistral can deliver a project. Thousands of agents used every day across the same company would show that Mistral has become difficult to replace.

Milestones that would make Mistral a real peer

Q14How big is Mistral vs. OpenAI or Anthropic?

Partly true: Mistral is a major AI company, but it is currently far smaller than OpenAI and Anthropic.

Financially, the comparison is one-sided. Mistral represents a low-single-digit share of either rival’s valuation and an even smaller share of Anthropic’s revenue. It has less capital, less computing capacity, fewer disclosed customers and no consumer product remotely close to ChatGPT.

The model gap is real too. Mistral can support useful business systems, but GPT and Claude currently perform much better on the hardest reasoning, coding and agentic tasks.

Mistral still carries more strategic weight than its size suggests. It offers downloadable models, private deployment and deep customization at a time when governments and large companies are nervous about depending on a handful of US providers.

That gives Mistral a credible path to becoming the leading supplier of controlled, customizable and European enterprise AI. It does not make the company a global peer yet.

Our judgment is direct: Mistral is one of the world’s most important AI challengers, but today it remains a specialist competing against two giants.