Mistral Vibe

Mistral Vibe is Mistral AI’s unified work-and-code assistant. In May 2026 Mistral folded Le Chat and its coding-agent line into one Vibe family spanning ordinary chat, web search, Deep Research, voice, images, documents, persistent work context, more than 100 connectors, long-running Work tasks and coding agents available on the web, in IDEs and from the terminal.¹²

The product is differentiated less by consumer scale than by architecture. Enterprise customers can use Mistral Cloud, private-cloud or on-premises deployment, with regional/data-residency options and custom models.¹ The current default model is Mistral Medium 3.5, a dense 128B open-weight multimodal model with a 256k-token context window.³⁴

Work mode is built for multi-step activity across connected systems. Mistral describes workflows that catch up across inboxes and calendars, perform deep research over public and internal sources, analyze spreadsheets, draft reports and coordinate recurring work.²¹

The connector layer includes email, calendars, Slack, GitHub, Jira and other services, with MCP compatibility.¹ Mistral’s Google connector documentation is unusually specific: Vibe can search and read Gmail, create drafts, access Calendar, create or update events, search Drive and read files. Write operations require explicit user approval.⁵

TechRadar’s multi-week hands-on review supports the product claim: Work Mode handled multi-step research and produced structured outputs in one run.⁶

Vibe for Code supports terminal use, VS Code and JetBrains, background cloud agents and a “teleport” workflow between remote and local environments.¹⁴ TechRadar reports a GitHub-connected workflow in which Vibe wrote and tested code and prepared a draft pull request, although the reviewer intervened twice with additional direction.⁶

The Register adds context around Leanstral, a proof-oriented coding agent inside Vibe. The publication reports Mistral’s own claims of strong formal-verification performance at relatively low cost; those figures remain vendor benchmarks rather than independent validation.⁷

Artificial Analysis reports Mistral Medium 3.5 at 30 on its Intelligence Index, around 140 output tokens per second and a 256k context window.⁸ LM Arena’s August 19 official dataset places Medium 3.5 around overall rank 41 with roughly 6,000 votes.⁹

The official Hugging Face model card confirms the 128B open-weight architecture and 256k context and reports Mistral’s own 77.6% SWE-bench Verified result.¹⁰ These signals support serious capability, but they do not place Medium 3.5 in the top current frontier cluster.

Mistral’s August Agentic Search release reports large gains from iterative retrieval and navigation on FinanceBench and OfficeQA Pro when Medium 3.5 is paired with Mistral’s search toolkit.¹¹ Those are first-party results, but they illustrate why the finished Vibe product should not be judged only by a base-model leaderboard.

Free costs $0 and includes web/mobile access, limited messages, searches and coding sessions, image generation, more than 100 connectors and $10/month in API credit. Pro is $14.99/month before tax and adds more usage, complex long-running tasks, all-day coding, more images, $30/month in API credit and support. Team is $24.99/user/month before tax and adds collaborative workspace features, storage, domain verification and data export. Enterprise is custom-priced and adds private deployment, custom models/agents/workflows, audit logs, SAML SSO and white labeling.¹²

The Pro price is below the conventional $20 individual premium tier of several larger rivals. The weakness is quota transparency: public pricing uses relative language such as “limited,” “more” and “up to 5x” rather than a stable request budget, an issue also noted by TechRadar.¹²⁶

API economics are separate. Medium 3.5 is currently listed at $1.50 per million input tokens and $7.50 per million output tokens.³

Google Play provides the largest current mobile sample, around 4.1/5 from roughly 43,900 reviews and more than one million downloads.¹³ Positive comments describe Vibe as useful and inexpensive; negative themes include instruction-following failures, latency, restrictive limits, connector problems, support delays and interface changes.

The directly readable Apple App Store sample is much smaller, while G2 currently has only one verified-current-user review.¹⁴¹⁵ These surfaces confirm real usage but are too uneven in sample size to create a highly stable professional-user consensus.

Trustpilot is around 2.5/5 from 96 reviews, with complaints about answer quality, limits, outages, connectors, billing and support, alongside positive comments about coding and the European alternative.¹⁶ Because the population is self-selected and service-heavy, it is most useful as qualitative warning evidence.

Mistral states that Free Vibe conversations may be used to improve models unless the user opts out. Pro, Team and Enterprise conversations are not used for model training by default; API data is not used for model training; uploaded Library documents are also excluded.¹⁷

Normal connector data is fetched on demand and is not permanently stored, and connector content is not used for training.¹⁷ There is an important transition caveat: the older Google Drive/SharePoint Knowledge Connector architecture indexes and stores files in European data centers and is being deprecated in favor of MCP connectors.¹⁸

Optional Memories can retain preferences across conversations, but users can view, edit, delete or disable them.¹⁹ Enterprise deployment can run on premises, in private cloud or on Mistral Cloud with data-residency controls.¹

Write actions on connectors such as Google Workspace require explicit approval.⁵ That reduces the chance that an agent silently changes external systems. It does not eliminate ordinary model error, and user evidence still includes complaints about instruction following and reliability.¹³¹⁶

Mistral’s offline-model documentation also warns users to validate outputs on representative tasks and to understand that external MCP tools or connectors may still send data outside the local machine.^{20 21}

Vibe is most compelling for professionals, developers and organizations that want a capable general assistant plus coding and agentic work without automatically defaulting to a U.S. hyperscaler ecosystem. It is especially relevant where European deployment, open weights, private infrastructure, data residency or model-training defaults matter.

The $14.99 Pro tier makes serious experimentation inexpensive. Free is useful for orientation but intentionally constrained. Team is mainly an organizational/admin step rather than a guaranteed model-quality upgrade, while Enterprise is where Mistral’s sovereignty and private-deployment proposition becomes most differentiated.

Vibe is a weaker fit for buyers who want the strongest currently benchmarked model regardless of price, teams whose workflows depend primarily on native Microsoft 365 or Google Workspace integration, or organizations that require a very large independent customer-review corpus before procurement.