Outlier vs Mistral Le Chat (now Vibe): local Mac AI vs a European cloud assistant

Full disclosure: I built Outlier and I sell it, which is an obvious bias. So I've written this from Mistral's own pricing page, docs, and privacy policy, and linked it all at the bottom so you can check me. Mistral's assistant does several things Outlier simply cannot.

What Mistral Le Chat is

Le Chat is Mistral AI's consumer and business assistant. It has been rebranded to Mistral Vibe: Mistral's App Store listing is titled "Vibe by Mistral (ex-Le Chat)" and says "Le Chat is now Vibe," and the pricing page says "Vibe (or Le Chat)." The entry point is still chat.mistral.ai.

Vibe now runs in two modes, Work (quick conversations and multi-step professional tasks, merged from the old Chat and Work tabs) and Code for development, available "across web, mobile, your code editor, and your terminal." That means a web app, iOS and Android apps, a terminal CLI, and VS Code and JetBrains plugins. There's no desktop app on any platform, including macOS; the pricing page's Interfaces list covers web, CLI, IDE, and mobile only. Desktop "Le Chat" wrappers on GitHub are community projects, not Mistral software.

Mistral's model card says Mistral Medium 3.5, a 128B dense model with a 256k context window, replaced its previous chat models in Le Chat and Devstral 2 in the Vibe coding agent. It's cloud-served, but the weights are published, which is unusual.

What Outlier is

Outlier is a native Mac app for Apple Silicon (M1 or later, macOS 15+). It downloads a model once, then runs it on your machine: no account, no caps, and it works with the Wi-Fi off. Six tiers, from Nano (4B, 2.37 GB, runs in 6 GB of RAM) up to Plus 397B-a17b, which streams experts from SSD and needs a 64 GB machine.

Nano and Lite are free, no account. Pro covers all six tiers, the Outlier Lifetime Licence at $49. What Outlier does not have, plainly: image generation, a mobile app, a hands-free voice mode, Windows or Linux builds, cloud sync. Web search and Deep Research do exist, but only on Pro and only when you turn them on.

Side-by-side comparison

Where Mistral genuinely wins

Several of these are things Outlier structurally can't do.

• Web search. Vibe searches the live web on every tier. Outlier's free tiers can't reach the web at all, and Pro's opt-in web research runs on DuckDuckGo, Mojeek and Wikipedia, a far smaller retrieval stack that is off until you switch it on. If your work depends on this week's news, that gap decides it.
• Image generation, voice, and mobile. Image generation is included even on the free tier. Voice mode handles recording, transcription, and summaries. There are real iOS and Android apps. Outlier can attach images and transcribe speech on Pro, but it generates no images, has no hands-free voice mode, and runs only on a Mac.
• Price. Pro is $14.99/month, and Outlier is not a subscription at all. Verified students pay $5.99/month for up to 12 months.
• The weights are actually open. Mistral Large 3 (675B) and Devstral Small 2 (24B) are Apache-2.0 on Hugging Face; Medium 3.5 (128B) is under a modified MIT license. If Mistral vanished, the weights would still exist.
• The CLI is real open source and isn't cloud-locked. github.com/mistralai/mistral-vibe is Apache-2.0, about 5,150 stars, actively pushed as of 2026-10-19. An account is optional, and the docs say it "works fully offline with local models, or against any compatible API key you provide."
• Ecosystem depth. 100+ connectors, custom remote MCP servers, Workflows, scheduled tasks, hooks, a code-interpreter sandbox, 15 GB storage on Pro.
• Agents that work with the lid closed. "Teleport" hands a coding session from your terminal to a remote agent and back. When your Mac sleeps, Outlier stops.
• EU-native posture. EU providers prioritized, Standard Contractual Clauses for transfers, self-serve deletion, a training opt-out, and on-premises or private-cloud deployment for enterprise.
• Coding scores. Mistral Medium 3.5, which replaced Devstral 2 in Vibe, reports 77.6% on SWE-Bench Verified. On a blind slice of that benchmark Outlier's local 27B measured 46% (23/50, build 1.11.757, 2026-08-09). The cloud model is ahead.

What it costs over a year

Mistral doesn't publish an annual plan, so the honest arithmetic is twelve monthly payments: $14.99 × 12 = $179.88 a year, before tax. The App Store listing shows a second in-app purchase at $149.99 alongside the $14.99 one, consistent with annual billing, but it doesn't label the period, so treat that as an inference, not a price. Outlier Pro is a one-time purchase, the Outlier Lifetime Licence at $49. Mistral Team is $24.99/user/month with a $50/month minimum, so a solo user on Team pays for two seats.

The bigger caveat: you can't tell what you're buying. Free is "limited messages and web searches." Paid is "Up to 6x free" messages and "Up to 5x free" web searches, every tier marked "subject to fair usage limits." No absolute numbers are published. Vibe Code Web is the one place with hard figures: 2 sessions per day free, 100 paid, 24-hour maximum session, 3-hour inactivity timeout. Outlier has no caps: no server to meter, so the ceiling is your Mac.

What happens to your data

This is where the two differ most. Read the policy, not the marketing.

Training is opt-out, not opt-in. The policy covers "Your Input and Output, subject to your opt-out," and the only scope carve-out is for people processing personal data in the context of their business activities, where Mistral acts as processor. A Pro subscriber using it personally isn't covered by that. An account-level control to object exists: you have to find it and flip it.

Retention has no default expiry. Mistral keeps input and output "until you delete your account or until you delete the conversation." There's no consumer equivalent of the API's zero-data-retention mode.

"European AI" doesn't mean data never leaves Europe. The policy prioritizes EU providers but reserves the right, "in exceptional cases," to use non-EU ones under Article 46 safeguards: a well-documented position, but not a guarantee. A Memory feature, now becoming Knowledge Base, also stores past interactions; the policy says Knowledge Base "will not seek to store sensitive data included in your Input," and relies on your explicit consent for any sensitive data that ends up in Memories.

Outlier's answer is structural, not contractual: no server, so no retention policy, no training toggle, no transfer mechanism. That's not virtue on my part. It's what happens when inference runs on your own silicon. And one wrinkle on Mistral's side: only the CLI is open source, so the web and mobile clients most people touch have to be taken on trust.

Who should pick which

Pick Mistral Vibe if you need live web results, image generation, voice, or a phone app; you want one $14.99 subscription covering chat and coding; you're a student eligible for $5.99; you want agents that run while your laptop is shut; or you're an EU organization that wants a GDPR-native vendor with an on-premises path.

Pick Outlier if you handle material you can't upload (client files, patient notes, unreleased code, legal drafts) and want that guaranteed by architecture, not a setting; you work offline; or you'd rather pay once with nothing metered. Cloud models like Claude are still faster and stronger on the hardest work; we published a 54-prompt side-by-side of Core 27B and Claude Opus so you can compare the answers yourself.

Run both if you can. Vibe for the live web or a phone; Outlier for sensitive, offline, or high-volume work. That's how I use it.