Check the current free-tier situation before you build on it. Mistral AI is a French lab that sells open-weight and hosted models through Mistral Studio (the API console), the Vibe assistant and enterprise deployments. The usage docs describe a Free mode that lets you create API keys and use included monthly usage within per-model limits shown in the Admin Panel, and paid use runs on prepaid pay-as-you-go credits. In October 2026 users reported that free organizations saw a banner saying to upgrade to activate API keys, and Reddit threads say the free API tier for direct access has been removed or paused for over a month. The docs I read had not changed. G2 shows 4.2 from 57 reviews, and most recent ones are incentivized. Reviewers like the price, multilingual quality and open-weight options, and complain about weaker results on hard tasks and limits that appear once usage grows.
Free tier quotas
Free mode (per the docs)
Create API keys and use included monthly usage within the limits on the Admin Panel Limits page
Published limits
Not published. A help article dated August 12, 2026 sends you to the Limits page for live figures
Status reported in October 2026
Users on free organizations report a banner saying to upgrade to activate API keys, and 429 errors on Medium and Small models
Earlier user-reported Experiment plan
1B tokens a month, 500K tokens a minute and 1 request a second, from a January 2026 GitHub issue
Free Vibe plan
Limited messages, web searches and coding sessions, basic image generation and 100+ connectors
Mistral Studio
Included with the Free plan per the pricing FAQ
Paid API use
Pay-as-you-go draws from prepaid credits; Tier 2 starts at 20 euros billed and Tier 3 at 100 euros
Batch and caching discounts
Batch processing 50% off; cached input tokens up to 90% off
Self-hosting
Open-weight models such as Mistral 7B are Apache 2.0 for research and individual use; commercial deployments need a Mistral license
Last sniffed Oct 7, 2026. Always confirm on the official pricing page.
👍 Good boy
+G2 rates it 4.2 from 57 reviews, with 54% giving 5 stars and 38% giving 4 stars. Most of the recent reviews on that page carry an incentivized (G2 invite) label, so weigh them accordingly.
+A 5-star G2 review is titled Cheap, fast and supports agentic caching, and the reviewer's main complaint was that even the large model sometimes was not clever enough for their specific tasks.
+A 4-star G2 reviewer said the console makes it easy to jump between model tiers, from the Ministral models up to Mistral Large, and that the smaller and mid-tier models are competitive on cost against much larger models from other providers.
+The same review is titled Easy Model Tier Switching and Strong Multilingual Performance at Great Value, so multilingual quality is part of what that reviewer rated.
+Open weights give you an exit. A G2 reviewer said they can use the models through the API or deploy open models themselves when they need more control over data and infrastructure.
+The pricing page spells out the cost levers: batch processing is 50% cheaper and cached input tokens cost up to 90% less, with OCR priced per 1,000 pages and speech models per minute.
+Studio gives usage breakdowns by model, input and output tokens, and workspace, and admins can set monthly spending limits per organization and per workspace.
👎 Watch out
−The free API tier is unstable right now. A GitHub discussion from October 10 to 16, 2026 shows the console banner saying to upgrade to activate API keys, 429 errors on Medium and Small models, and Large returning a tier-not-allowed error while Ministral and Codestral still answered.
−A Reddit thread titled Are we going to ignore that mistral removed free tier? says the free tier has not been restored after over a month and was removed from the status page. A comment quoted in the search excerpt says the free tier for direct API access is being removed, but I could not confirm who wrote it.
−Mistral does not publish free-mode rate limits. A third-party tracker and Mistral's own help article both point to the Admin Panel Limits page, so you only learn your numbers after signing up.
−Quality gaps show up in G2 reviews. One reviewer said even the large model is sometimes not clever enough for specific tasks, and another listed occasional hallucinations, answers given with high confidence, and poor coding compared with some top models.
−A G2 reviewer said the ecosystem and documentation need more hands-on effort than on the more established AI platforms, and that API rate limits and usage constraints become noticeable as usage grows.
−A G2 reviewer called the free offer quite limited, said daily use pushes you to a paid plan, and said native integrations with tools like Notion and Google Workspace are missing.
−Third-party reports from September 2026 say Vibe free-tier conversations feed model training unless you opt out in the settings. Check data settings before sending anything sensitive.
The catch
The docs describe a Free mode: you create API keys and use included monthly usage within the limits on the Admin Panel Limits page. Mistral does not publish those limits. A help article dated August 12, 2026 says the Limits page shows your live numbers, and pay-as-you-go draws from prepaid credits. Higher billing tiers open as cumulative billed amounts grow, with 20 euros for Tier 2 and 100 euros for Tier 3, according to a third-party summary of that article.
In October 2026 that picture stopped matching what some users saw. A GitHub discussion between October 10 and 16 shows a free organization with a console banner reading Your organization is on the Free plan, Upgrade to activate your API keys and access the full API, with the New key button greyed out. Medium and Small returned 429, Large returned a tier-not-allowed error, and Ministral and Codestral still answered. Reddit threads say the free tier for direct API access was removed for over a month. The docs page I read was cached on October 21 and still described Free mode, so treat the free API as unconfirmed until you see keys activate on your own account.
Earlier in 2026 users described an Experiment plan with 1B tokens a month, 500K tokens a minute and a 1 request a second cap, with lower priority at peak times. That came from a user in a GitHub issue, not from Mistral.
I did not find statements about a credit card at sign-up or about commercial use of free-mode output, so both are unknown. The free Vibe assistant is a separate consumer product with limited messages and searches.
What’s included
✓Hosted API for Mistral models with token-based pricing
✓Open-weight models for self-hosting
✓Mistral Studio console with playground and usage dashboard
✓OCR priced per 1,000 pages
✓Voxtral speech models priced per minute
✓Batch processing at 50% off
✓Prompt caching with up to 90% off cached input
✓Organization and workspace monthly spending limits
✓Admin API on Enterprise plans
✓Vibe CLI and IDE coding agents
Free tier value
Generosity
B
Estimated value
Free mode documented but limits not public and availability disputed; Pro lists $15 in credits
Feature parity
Pay-as-you-go lifts limits by billing tier. Model access also depends on tier.
Screenshots
HomepagePricing page
Paid plans
Pay-as-you-go API credits
Extra usage beyond the included amount draws from prepaid credits. Most models are priced per million input and output tokens. The FAQ example lists Mistral Large at 0.50 dollars in and 1.50 dollars out per million tokens.
Pro, 14.99 dollars a month
Higher chat limits and full access to Vibe for coding. The plan lists 15 dollars a month in API credits. Verified students get Pro for 5.99 dollars a month.
Team, 24.99 dollars per user a month
Higher limits than Pro, up to 30 GB of storage per user, domain verification and data export.
Enterprise, contact sales
Private deployments, custom models and agents, audit logs, SAML SSO, white labeling and the Admin API.
Common questions
Does Mistral have a free API?
The docs describe a Free mode with included monthly usage. In October 2026 users reported blocked keys and 429 errors on free organizations, so check your own console.
What are the free rate limits?
Mistral does not publish them. The Admin Panel Limits page shows your numbers. A January 2026 GitHub issue described 1 request a second on the Experiment plan.
How do I raise limits?
Turn on pay-as-you-go. A third-party summary of the help article says billing tiers rise with cumulative billed amounts, and that adding credits alone does not raise them.
Can I self-host Mistral models?
Yes. The pricing FAQ says open-weight models such as Mistral 7B are Apache 2.0 licensed for research and individual use, while commercial deployments need a Mistral license.
Does Mistral train on my data?
Third-party reports from September 2026 say Vibe free-tier chats train models unless you opt out. Check the data settings for both Vibe and the API.
What do G2 reviewers dislike?
Weaker results on hard tasks, occasional hallucinations, a thinner ecosystem and documentation than established platforms, and rate limits that show up as usage grows.
Change history
Checked October 22, 2026
Docs still describe Free mode. GitHub and Reddit threads from October 2026 report free organizations blocked from API keys. G2 shows 4.2 from 57 reviews.
