Perplexity against Mistral: the engine that cites its sources against the assistant that protects yours

Perplexity promises a sourced answer to any question, in seconds. Mistral, through its Le Chat assistant, promises a general-purpose AI « made in France », with your data staying on European soil. Both promises are true. Neither is the whole story. Here is the match as of 4 July 2026 — figures, lawsuits, security holes and real limits included, without the marketing varnish most comparisons simply copy over.

---

The verdict, in short

Across the seven criteria in this comparison, Mistral wins 4-1, with two ties — a clear advantage, but not a rout. Mistral wins because it costs less from the free tier upwards, hosts its data in Europe, and faces, to date, no publisher lawsuit and no documented critical security hole. Perplexity is not beaten everywhere: on its home ground — real-time sourced search — it remains the reference, and that is precisely what it was built for.

---

Two stories, two countries, two speeds

Perplexity was born in August 2022 in San Francisco, founded by Aravind Srinivas (ex-OpenAI, ex-Google DeepMind) with Denis Yarats, Johnny Ho and Andy Konwinski. In a little over three years the company went from 10 million monthly active users (early 2024) to 45 million (end of 2025) — 800% growth in a year. Its valuation followed: $20 billion in September 2025, then $21 to $22 billion in early 2026 after a new funding round. Among the investors: Nvidia, Jeff Bezos personally, SoftBank, and since December 2025, Cristiano Ronaldo. Annualised revenue was running at around $150 to $200 million at the end of 2025, with an internal target of $656 million for 2026. Aravind Srinivas has himself floated an IPO around 2028.

Mistral AI, which publishes Le Chat, was founded in Paris in April 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix, three researchers out of DeepMind and Meta. Its financial trajectory is almost as vertiginous: valuation doubled in a year, from 5.8 to 11.7 billion euros after raising 1.7 billion euros in September 2025, led by — of all companies — ASML, the Dutch chip-lithography giant, which now holds around 11% of the capital. Mistral’s annualised revenue went from $20 million in early 2025 to more than $400 million in January 2026, with a stated target of a billion by the end of the year.

Two young companies, two funding success stories — but two product philosophies with, fundamentally, almost nothing in common.

What each one actually does

Perplexity is not a general-purpose chatbot: it is an answer engine. Its speciality, since it launched in December 2022, is to search the web in real time and cite its sources precisely — a sort of anti-ChatGPT designed to replace Google search rather than conversation. Around that core it has built an impressive stack: the Comet browser (free since March 2026 on iOS, Android, Windows and macOS), able to browse and act on your behalf; Perplexity Labs for generating reports and spreadsheets; and, from the $200/month Max plan, « Perplexity Computer », which orchestrates 19 different AI models as specialised sub-agents depending on the task. Perplexity Pro also gives access, under a single subscription, to GPT-5.2, Claude Sonnet 4.6, Gemini 3 Pro, Grok 4.1 and Kimi K2 alongside its own Sonar model — a genuine argument for anyone who wants to compare AI models without multiplying subscriptions.

Mistral, through Le Chat, is a different bet: a French general-purpose AI positioned on confidentiality and professional versatility rather than pure search. Renamed « Vibe » for its agentic side, the tool combines a work mode (inbox management, search, document writing, connected to more than 100 tools including Slack, GitHub and Jira via the MCP protocol) and a code mode, powered by Devstral 2 and the flagship Mistral Medium 3.5 (128 billion parameters, a 256,000 token context window). On SWE-bench Verified, the industry reference for code generation, Devstral 2 reaches 72.2%. Mistral has also launched Voxtral, its own speech model, claimed to be 73% cheaper than ElevenLabs. Notably, and rarely in this industry: most Mistral models (Large 3, and the small dense models alongside it) are published as open weights under the Apache 2.0 licence — anyone can download them, inspect them and run them locally. Perplexity, by contrast, publishes no open model: Sonar stays proprietary, and the rest of the offer consists of reselling access to other people’s models.

The price, without the small print

Mistral’s free offer is markedly more generous than most competitors’, which restrict their flagship model on the free tier: here it is the most capable model in the catalogue that is served without condition, quota aside. Perplexity offers no equivalent to Mistral’s student rate.

On pure price and free-tier generosity, Mistral wins unambiguously. On the depth of the paid offer — multi-model access, an agentic browser, orchestration of 19 models — Perplexity Max has no real equivalent, provided you are willing to put $200 a month into it.

The angle most comparisons leave out: trust

This is where this comparison parts company with most of the specialist coverage on the subject, which mostly compares speed and answer quality. On actual trust — the kind measured in lawsuits, security holes and customer reviews — the picture is considerably less flattering for Perplexity.

Since 2025, Perplexity has taken a wave of lawsuits from the US press: Dow Jones and the New York Post (News Corp) accuse it of scraping their articles without permission, and a federal court in New York refused to dismiss the case. The New York Times filed its own infringement claim in December 2025. CNN followed in May 2026, with a detailed complaint over the taking of more than 17,000 articles, videos and images. Cloudflare, separately, accuses Perplexity of circumventing robots.txt instructions by disguising the origin of its traffic to pass for an ordinary browser — a practice the company denies.

The most spectacular conflict pits Perplexity against Amazon. In November 2025, Amazon sued Perplexity, arguing that the shopping agent built into Comet was accessing password-protected customer accounts by passing itself off as Google Chrome, despite repeated warnings. On 9 March 2026, a federal judge sided with Amazon and blocked Comet on its platform by preliminary injunction, holding that permission given by a user to their AI agent is not permission from the platform itself — a legal distinction that could set precedent for all agentic commerce. Perplexity has appealed to the Ninth Circuit.

On technical security, Comet has had two documented critical vulnerabilities: « CometJacking », which allowed emails, calendar and credentials to be stolen through a single click on a booby-trapped link, with no further interaction; and « PleaseFix », which allowed the agent to be hijacked remotely and local files accessed through a malicious calendar invitation, the agent continuing to answer normally throughout the exfiltration. Reported on 22 October 2025, that second hole was only fixed on 23 January 2026 — close to three months of exposure on a browser already publicly deployed. Add a Trustpilot rating of around 1.6/5 across nearly 180 reviews, dominated by complaints about surprise billing and impossible cancellation, and the picture becomes seriously more nuanced. In Perplexity’s favour: the company announced in February 2026 that it was permanently dropping advertising from its answers, judging it incompatible with user trust — a rare and rather virtuous choice in this industry.

Mistral is not free of grey areas either. A complaint has been filed with the CNIL, France’s data protection authority, over Mistral making the right to object to data processing conditional on a paid subscription for free-tier users — a practice potentially at odds with article 12 of the GDPR, which forbids charging for the exercise of that right. Since the complaint, Mistral has opened an opt-out route by email for free accounts, more laborious than the one-click button paying subscribers get; the CNIL has not yet ruled. Security researchers also showed in early 2026 that jailbreak techniques could make its flagship model produce pro-Russian propaganda — a documented weakness, though not specific to Mistral, which affects most large models according to that same work. No critical security hole comparable to Comet’s, and no publisher lawsuit, has been documented against Mistral to date.

Sovereignty and compliance: the real dividing line

This is probably the most structural difference between the two products. Mistral’s Pro and Enterprise plans guarantee 100% data residency within the European Union; Mistral is building its own compute centre in Essonne, France (13,800 Nvidia GB300 GPUs) and has just announced, with Bpifrance and MGX, an AI campus project targeting up to 3 GW of capacity in France. The French state has rolled out an assistant based on Mistral models to more than a million public servants out of 5.7 million, for a total cost of around 750,000 euros — a large-scale deployment that amounts, in practice, to a vote of confidence at the highest level, even if the opening to local authorities is not planned until 2027. The European AI Act, whose key milestone takes effect on 2 August 2026, is a structural advantage for Mistral, whose models were designed from the outset to meet European transparency and traceability requirements.

Perplexity, a US company, offers no equivalent guarantee of data residency in Europe and remains subject to US law — a point rarely highlighted in comparisons that concentrate on response speed.

The real limits, not hidden

Mistral lags on the most complex reasoning tasks and the hardest technical problems, where Claude and GPT-5.5 keep the advantage according to several cross-checked benchmarks. It has fewer third-party integrations than ChatGPT or Gemini, which can limit it inside workflows already built around other ecosystems. Perplexity, for its part, is simply not a coding or office tool: it is a search engine with an agentic browser attached, not a generalist. The two therefore meet head-on across a narrower set of grounds than it first appears — the rest depends on what you are trying to do, not on an absolute ranking.

One more under-documented limit: when an AI agent — whether Vibe’s work mode or Comet’s shopping agent — gets broad access to your tools, the question is no longer only « is the answer right? » but « what can it do without asking you? ». The Perplexity-Amazon legal battle is the most concrete demonstration of that to date: an agent acting « with the user’s permission » does not necessarily have the permission of the platform it is acting on, and that ambiguity is not specific to Perplexity.

You will find the same reflex applied across our AI comparisons and our tested promises: look at what the tool does when nobody is watching, not at what the landing page says.

---

What to remember

• Overall winner in this comparison: Mistral (4 criteria out of 7, two ties). Price, data sovereignty and the absence of publisher lawsuits tip the balance.
• Perplexity remains the reference for real-time sourced search — its original function, one Mistral does not claim to rival.
• On the free tier, the gap is wide: Mistral serves its full flagship model, Perplexity a lighter one.
• On the legal file, the gap is wider still: three publisher lawsuits and a lost injunction against Amazon on one side, an unresolved CNIL complaint on the other.
• Neither is a general-purpose winner. One searches and cites, the other works and protects. Pick by what you do, not by a score.