LATEST
AI Tools

Mistral Vibe Desktop Review: Parallel Agents, Local Files, and 3 Free Months in Europe

Updated November 19, 2026
16 min read
Editor's Note: This article was human-reviewed for accuracy
Developer typing code on a laptop while pointing at JavaScript on the screen

Quick Answer & Key Takeaways

  • • Mistral launched Vibe Desktop on November 17, 2026, a macOS and Windows app that runs several coding agents in parallel and reviews their diffs and pull requests in one window.
  • • Pro costs $14.99 a month before tax and includes $25.50 in monthly API credits, while developers in Europe get three months of Pro free with the code LECHONK.
  • • The catch is the model behind it: Mistral Large 4 leads open models on coding tests but still trails the best closed models by a wide margin.

Mistral shipped its coding agent as a desktop app on November 17, 2026. Vibe Desktop puts the company's Work and Code modes in one window for macOS and Windows, runs several coding agents at once, and lets you review their diffs and pull requests without leaving the app. Until now, Vibe Code ran in a terminal, in VS Code, or in the browser.

The launch comes with a sharp offer for one group of users: developers in Europe get three months of Vibe Pro free with the code LECHONK. For everyone else, the question is whether the app is worth $14.99 a month, and whether the model behind it is good enough to replace the coding assistant you already pay for. This review works through what the app does, what it costs with the fine print, and where it still falls short.

What launched on November 17

Vibe is the product Mistral used to call Le Chat. The company renamed its assistant on May 28, 2026, and turned it into an agent with three modes: Chat for conversation, Work for multi-step tasks across connected apps, and Code for developers. Vibe Code already ran in four places: the Vibe CLI in the terminal, a VS Code extension, the web at chat.mistral.ai, and remote cloud sessions. The desktop app is the fifth surface, and the first that puts Work and Code side by side in one window.

Mistral announced the app in a thread on X on November 17, 2026, and published release notes the same day in its documentation. The app installs on macOS and Windows from Mistral's Vibe product page, and you sign in with a Mistral account. Linux support is listed as coming soon, with no date attached. The download is free; what you can do inside it depends on your plan.

The pitch, in Mistral's own words from its launch announcement, is to hand Vibe several tasks at once, watch the agents work side by side, then review their diffs and track pull requests in the same window. Mistral pairs the app with its newest model, Mistral Large 4, which it nicknames Le Chonk, and calls the two together "the world's best open model-harness duo". That is the company's claim, not an independent finding, and the benchmark section below tests it.

What the app actually does

The desktop app is built around parallel work. Mistral's release notes describe three ways to split a job up. Git worktrees give each task its own checkout of the repository, usually on another branch, so two agents never edit the same files. Sub-sessions let the main session open nested conversations in the project's sidebar, each sending results or questions back to it. Subagents take on part of an investigation, and you read each subagent's conversation in a read-only view with a status of Running, Waiting for input, Idle or Failed. Two conversations on the same checkout share the same files, so Mistral's own documentation warns to ask Vibe for a new worktree for each task when two agents touch the same repository.

Every conversation runs in one of two environments. A local session works against a repository on your own machine and in your own development environment, running commands in the integrated terminal. It shares the Vibe CLI's core agent engine and session logs, so no separate CLI installation is needed, though Mistral notes that tools and prompts can differ between the app and the CLI. A cloud session runs the agent in an isolated sandbox against a GitHub repository connected to a project, the same setup as Vibe Code on the web. A task started on the laptop keeps running in the cloud after the lid is closed, according to Mistral's launch video, which makes the cloud sessions the right choice for long jobs you do not want to babysit.

Review happens inside the app. The Git Changes panel shows each file's diff in a unified or split view. You can run your own tests in the built-in terminal and keep documentation open in the integrated browser while a session runs. A merge icon at the top right lists the session's pull requests with their number and status, including Conflicts. For a pull request with conflicts, Mistral's documentation suggests asking the agent to resolve them and rerun the tests, then reviewing the result before merging. Nothing in the app merges for you on its own.

Work mode gets local context too. You can add local folders to Vibe Work alongside its connected tools and shared Work capabilities, which include more than 100 connectors for email, calendar, Slack, GitHub and Jira. Background commands keep builds, development servers and file watchers running while the agent works, with live output and a stop button in their own panel. Conversations can be grouped into projects, pinned or archived, and switched with a command palette and keyboard shortcuts. The app takes voice input as well, and Mistral says that input runs on Voxtral, its audio model. Our review of Mistral Voxtral 3 covers that model in detail.

Mistral's product page leads with a security example. A user asks how a red team pulled cloud credentials out of a document-preview service. Vibe explains the attack, writes a log query to confirm it, then patches the code, adds tests and opens a pull request. That is Mistral's own demo, not an independent test, and it points to the pitch for the model behind the app: security work that other agents refuse to touch.

The model behind it: Le Chonk by default

Paying users get Mistral Large 4 as the default model in the desktop app. Mistral confirmed that in its launch announcement on GitHub, and its product page leads with the pairing. Free users do not get the same default, and Mistral has not said publicly which model the free tier uses in the app.

Large 4, nicknamed Le Chonk, is Mistral's largest model: 1 trillion parameters with 52 billion active per token, natively multimodal, with a 1-million-token context window according to Mistral's model page. Mistral put it in public preview on October 6, 2026, then released the weights on October 29 under the Mistral Ecosystem Open-Weight License, a modified MIT licence that lets businesses use, modify, fine-tune and commercialise the model, with one condition: a business offering the model as a hosted service needs a separate commercial licence from Mistral first.

The API price changed on the same day the weights came out. Mistral's API now lists Large 4 at $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens. The preview list price was $1.36 input and $4.18 output, so the current rate is roughly half the preview price. The model is still labelled a public preview on Mistral's model page, so prices and limits can change.

On coding benchmarks, Large 4 leads the open models, but the lead is thin and the gap to closed models is wide. Every figure in this section is Mistral's own measurement unless stated otherwise.

BenchmarkMistral Large 4GLM-5.3DeepSeek V4.1 Flash
Terminal-Bench 4.041%40%31%
FrontierCode 2.019.7%17.8%19.3%

The 41% Terminal-Bench figure comes from the October 29 weights release; the preview scored 28.3% on the same test. On the preview announcement, Mistral also reported 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA, and a combined Coding Agent Index of 49.8%, ahead of DeepSeek V4 Pro and Qwen3.8 Max. A blind human evaluation run with Surge AI ranked Large 4 second of five models on coding quality, behind only Claude Opus 5.

Closed models sit far ahead on the same tests. On FrontierCode 2.0, Claude Fable 5.2 scores 49.1% by Anthropic's count, more than twice Large 4's 19.7%. Mistral's comparison is with open models, and its launch post says Vibe is "actually better than GLM / DeepSeek at coding". That is true on the numbers Mistral publishes. It is also true that the best closed coding model scores more than double Large 4 on the same benchmark, so the "best open model-harness duo" claim is a claim about open models only, and buyers who need the strongest coding model available should read it that way.

The security numbers are the strongest part of the model's case. On the Artificial Analysis Cyber Index, Large 4 ranks among the top five models globally. On one of the index's tests, reproducing a real vulnerability in open-source software and then patching it, the model scores 82%, the highest of any model, and it solves 93% of the challenges in Cybench. Mistral's announcement notes that several leading closed models, including Claude Opus 5.5 and GPT-6 Astra, score near zero on the same test because they refuse to perform the task. That refusal is a real difference: defending software often starts with proving a flaw is real, and safety filters in closed models can block exactly that work. Whether that matters to you depends on whether you do security work at all. For ordinary feature work, the coding scores above matter more.

What it costs

One subscription covers both Work and Code. Mistral's pricing page, read on November 19, 2026, lists four tiers. Prices exclude taxes, and Mistral lists no annual price for any Vibe plan. UK and Canadian figures use November 19, 2026 market rates of 1 USD = 0.7610 GBP and 1 USD = 1.4336 CAD. Mistral bills in USD, so the pound and dollar figures are approximations.

PlanPriceWhat it adds
Free$0 (£0 / CA$0)Limited messages, web searches and coding sessions; web and mobile; 100+ connectors; 5 scheduled tasks
Pro$14.99 a month (£11.41 / CA$21.49)All-day coding in the CLI, IDE or web; up to 6x the free messages, 5x web searches, 40x image generations; 15 GB library; unlimited scheduled tasks; $25.50 a month in API credits
Pro Student$5.99 a month (£4.56 / CA$8.59)Pro features for up to 12 months, for verified students who have never used Vibe or Le Chat; $15 a month in API credits instead of $25.50
Team$24.99 per user a month (£19.02 / CA$35.83)Higher limits; up to 30 GB of storage per user; domain verification; data export
EnterpriseCustomPrivate deployments, custom models and agents, audit logs, SAML SSO, white labelling

The $25.50 of monthly API credits is the part of the price most buyers will miss, and it changes the arithmetic. Pro costs $14.99 a month and includes $25.50 a month in credits for Mistral's API, spendable in Mistral Studio. At Large 4's rates, that allowance buys about 36 million input tokens or about 12 million output tokens a month. The credits are worth nothing to a buyer who never makes an API call, and Mistral publishes no expiry or rollover rule for them, so treat them as a monthly allowance that resets, not as money in a bank.

Four pieces of fine print matter before anyone budgets around the subscription. First, the allowance is shared. Mistral's documentation says each plan includes one monthly allowance shared across Vibe Code, Studio and the API, spent before any prepaid credits. Heavy coding sessions and API calls draw from the same pool. Second, pay-as-you-go is off by default, and Vibe Code can stop when the allowance runs out, so a long agent session can halt mid-task until the month resets or you buy more credits at API rates. Third, Mistral states its Pro limits as multiples of the free plan instead of fixed counts, so you cannot work out in advance how many messages or coding sessions a plan buys; the numbers above are the multiples Mistral publishes. Fourth, the pricing table lists a model-training opt-out for Pro, Team and Enterprise, and lists none for the free tier. If your code must not be used for training, the free plan is not the way to find out.

The launch offer is the cheapest way in. Mistral is giving developers in Europe three months of Vibe Pro free with the code LECHONK, signed up through the link in its launch thread on X. The company calls the offer limited and first come, first served, and has not given an end date. Launch coverage reports the first batch of codes ran out on day one and Mistral added more on November 18, so the offer can disappear without warning.

How to get started

For the desktop app, download Vibe Desktop from mistral.ai/products/vibe, install it on macOS or Windows, and sign in with your Mistral account. Open a repository for a local session, or connect a GitHub repository for a cloud session, then describe the task in plain language. Start with one focused job, such as fixing a failing test or adding a small feature, before handing the app several tasks at once.

For the terminal, install the Vibe CLI with curl -LsSf https://mistral.ai/vibe/install.sh | bash on macOS or Linux, or with uv tool install --python 3.14 mistral-vibe on Windows. Run vibe in your project directory and sign in with your Mistral account or an API key. The CLI is the only surface with a non-interactive mode, vibe --prompt, which is the one to use for scripts and CI. The VS Code extension is the right choice if you want the agent next to your editor, with the active file and selection in context.

Two settings deserve attention on first use. Vibe Code asks for approval before running a shell command by default, but it applies file edits without asking. If you want to review edits too, pick the ask agent mode. And when two tasks touch the same repository, give each one its own Git worktree, or the agents will edit the same files and their changes will collide.

How it compares

Vibe Desktop arrives in a crowded field. Claude Code is the reference point for terminal coding agents, and our Claude Code explainer covers it in full. OpenAI's Codex is the other obvious alternative, bundled with ChatGPT. The table below compares the entry paid tier of each, with prices read from the vendors' own pages on November 19, 2026.

Mistral Vibe DesktopClaude CodeCodex
Entry paid price$14.99 a month (£11.41 / CA$21.49)$20 a month, or $17 a month billed annually ($200 up front)$20 a month via ChatGPT Plus (£15.22 / CA$28.67)
Default modelMistral Large 4, open-weightClaude Fable, closedGPT-6.2, closed
API credits included$25.50 a monthNone on Pro; $100 a month on Max ($100 a month)None on Plus
Free tierYes, with limited coding sessionsYes, but Code needs a paid planPaid plans only
Open-source agentYes, CLI under Apache 2.0NoNo

On price alone, Vibe Pro is the lowest-priced paid tier in this comparison, and the only one of the three that bundles API credits at the entry tier. The $25.50 credit is worth more than the $14.99 subscription on paper, which is the strongest argument for the plan if you also use Mistral's API. Claude's counter is the model: Fable scores 49.1% on FrontierCode 2.0 against Large 4's 19.7%, and Claude Code is included in every paid Claude plan, with Max adding $100 a month in API credits at $100 a month. Codex's counter is the OpenAI ecosystem and its computer-use models, which Vibe Desktop does not match.

The catches

Linux is the first gap. The app runs on macOS and Windows only, and Mistral says Linux is coming soon without a date. Developers on Linux keep the CLI, which is where Vibe started, but they do not get the parallel-agent window or the in-app review flow.

Computer use is the second. Rival agents are moving toward controlling the whole operating system, reading the screen and taking actions across apps. Vibe Desktop does not drive other applications on your computer. It works inside its own window, its terminal and its browser panel. That is a deliberate scope, and it keeps the app simpler to reason about, but it is also a real limit next to agents that can operate the machine itself.

The model gap is the third, and the one that matters most. Large 4 leads open models on coding tests by a point or two, and trails the best closed models by a wide margin on the same tests. If your work is security research, the open weights and the model's willingness to do vulnerability work are a genuine advantage. If your work is everyday feature coding, the closed models will usually produce better code, and the $5 a month Vibe saves against Claude Pro or ChatGPT Plus is small next to an engineer's hourly rate.

The free tier has its own limits. Coding sessions are limited, and Mistral reserves remote coding agents for Pro, Team and Enterprise plans. The LECHONK offer is Europe only, first come first served, with no published end date, and the first batch of codes already ran out once. Finally, this is a first release. The CLI underneath it is changing fast, with version 3.1 landing on November 9 and 10, 2026, adding hooks, a /pr command and an import of your setup from other coding agents. Expect the desktop app's behaviour to change with it.

Frequently asked questions

Is Mistral Vibe Desktop free?

The app is a free download, and the free plan includes limited messages, web searches and coding sessions. All-day coding, higher limits and remote coding agents need Pro at $14.99 a month, and developers in Europe can get three months of Pro free with the code LECHONK while places last.

Does Vibe Desktop work on Linux?

Not yet. The app runs on macOS and Windows, and Mistral says Linux support is coming soon with no date. Linux users can run the Vibe CLI instead, which installs with a single shell command and shares the same agent engine.

What model does Vibe Desktop use?

Paying users get Mistral Large 4, nicknamed Le Chonk, as the default model. It is a 1-trillion-parameter open-weight model with 52 billion active parameters per token. Mistral has not said which model the free tier uses by default in the app.

How do I get three months of Vibe Pro free?

Sign up through the link in Mistral's November 17 launch thread on X and enter the code LECHONK. The offer is for developers in Europe, first come first served, with no published end date. The first batch of codes ran out on launch day and Mistral added more on November 18.

Is Mistral Vibe open source?

The CLI is. Mistral publishes its source on GitHub under the Apache 2.0 licence, where it has more than 5,400 stars, though the project does not accept pull requests. The desktop app is a free download from Mistral, and the model weights are open under Mistral's own open-weight licence, not an open-source licence.

How is Vibe Desktop different from the Vibe CLI?

The desktop app runs several local and cloud sessions side by side in one window, with in-app diff review and pull-request tracking. The CLI runs one session in the terminal and is the only surface with a non-interactive mode for scripts and CI. Both share the same agent engine and session logs.

The bottom line

Vibe Desktop is a solid first release that closes the gap between Mistral's coding agent and its rivals on workflow, while the model behind it still trails the best closed options. For developers in Europe, the three free months of Pro make it a no-cost trial worth taking. For everyone else, $14.99 a month plus $25.50 in API credits is the lowest-priced paid tier among the three agents compared here, and the open weights and open-source CLI are real advantages for teams that need to keep their code in-house. Just go in knowing what you are buying: the app is the strong half of the package, and the model is the part still catching up.

Sources

16 min read
32 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more