Mistral AI API review
by Mistral AILLM APIs
Per-token access to Mistral's own, mostly open-weight models, with a free mode, prepaid credits and optional EU or US endpoints.
- Free plan
- No card needed
- Self-serve
- Usage-based
- Public prices
What you're paying for
- Worth the price (main value)
- Compliance included
- Saves engineering time
Low per-token prices on small open-weight models, with EU or US processing you can choose yourself and spending caps per workspace.
The Mistral AI API sells Mistral's own models, from Ministral 3 3B at $0.10 per 1M tokens up to Mistral Medium and Large, plus OCR, audio, embeddings and moderation, through one API with an OpenAI-style request shape. Most current models are open-weight: the model pages list Ministral 3 and Small 4 under Apache 2.0, while the pricing FAQ says commercial deployments need a Mistral license. Teams that need processing kept in the EU or the US get regional endpoints at 10% above list price without a sales call, and the help center states SOC 2 Type II and ISO 27001, with the SOC 2 report available to clients on request.
- A tool-calling model at $0.10 per 1M tokensMinistral 3 3B is GA with a 256k context, function calling and structured outputs, at $0.10 per 1M input and output tokens; our 100M + 20M token scenario comes to $12 a month (computed).Evidence: Pricing page and Ministral 3 3B model page (2026-10-15)
- Batch, caching and spending capsBatch is 50% off list price, and cached prompt tokens are billed at 10% of the input price. Monthly spending limits can be set per Organization and per Workspace.Evidence: Pricing page; batch, prompt caching and billing docs (2026-10-15)
- EU or US processing as a self-serve optionRegional endpoints keep inference in EU and EFTA or in US data centers at 1.1x list price. They support function calling but not Agents, Batch or the Files API.Evidence: Regional inference docs (2026-10-15)
- Open weights and an OpenAI-style APIThe model pages list Ministral 3 and Small 4 weights under Apache 2.0 (the pricing FAQ says commercial deployments need a Mistral license), and the migration guide shows the OpenAI client pointed at the Mistral base URL.Evidence: Model pages; mistral.ai pricing FAQ; migration guide (2026-10-15)
- Best for
- Teams that want low-cost small models with tool calling, EU or US processing and spending caps, on a self-serve prepaid account.
- Good to know
- Prompts and outputs may be used for training unless you turn it off in the Admin settings, on Free mode and pay-as-you-go alike.
How Mistral AI API compares
Against 3 other products we reviewed in LLM APIs: what you pay for with each, the five access facts, the plan limits we recorded and each review's bottom line. The note under the table says how each one was checked.
| Metric | Mistral AI API | Anthropic API (Claude) | OpenAI API | Google Gemini API |
|---|---|---|---|---|
| What you're paying for | Worth the priceLow per-token prices on small open-weight models, with EU or US processing you can choose yourself and spending caps per workspace. | Worth the priceClaude models from Haiku 5.5 at $0.10 / $0.50 per 1M tokens up to Fable 5.1, with no training on API data and a BAA you can sign in the console. | Saves engineering timeOne API key for OpenAI's GPT-6 models and hosted tools, priced per token with published batch, flex and caching discounts. | Easy to adoptGoogle's Gemini models by the token in US dollars, with an ongoing free tier to start and batch, flex and caching discounts to scale. |
| Overall score | 7.4Best | 7.2 | 7.1 | 5.3 |
| Free option | Free plan | Free trial | –Not measured | Free plan |
| Card to start | No card needed | –Not measured | –Not measured | No card needed |
| Getting access | Self-serve | Self-serve | Self-serve | Self-serve |
| Starter priceCheapest paid plan | $0.10/1M input tokens | $0.10/1M input tokens | $0.10/1M input tokens | $0.25/1M input tokens |
| Pricing model | Usage-based | Usage-based | Usage-based | Usage-based |
| Pricing page | Public prices | Public prices | Public prices | Public prices |
| How you pay | Prepaid pay-as-you-go credits, optional auto recharge | Prepaid credits by card, expire after 1 year; invoicing through sales | Prepaid credits ($5 minimum), expire after 1 year | Prepaid credits ($5 minimum), expire after 12 months |
| Free option | Free mode, no card; allowance not stated publicly | One-time free credits for new users, amount not stated | No free allowance stated | Ongoing free tier on select models, no card; limits shown in AI Studio |
| Reference scenario (100M in + 20M out tokens/mo) | $12 on ministral-3b-2512, computed | $20 on claude-haiku-5-5, computed | $20 on gpt-6-luna, computed | $55 on gemini-3.1-flash-lite, computed |
| Batch, cache and other discounts | Batch 50% off; cached input 90% off | Batch 50% off; cache hits 90% off on Haiku 5.5 | Batch and Flex 50% off; cached input 90% off | Batch and Flex (preview) 50% off; cached input 90% off |
| Trained on API data | Unless you opt out (Free mode and paid) | No (Commercial Terms) | No, unless you opt in | Paid: no. Free tier: may be used to improve Google products |
| Default retention | 30 rolling days for abuse monitoring | Deleted within 30 days | Abuse logs up to 30 days | Abuse logs for a "limited period", days not stated |
| Zero data retention | By approval, pay-as-you-go, stateless endpoints only | By agreement, through sales | By approval, through sales | Not guaranteed: paid prompts are logged for abuse monitoring (docs point to Gemini Enterprise Agent Platform, formerly Vertex AI) |
| SOC 2 / ISO 27001 | Both, stated in the help center | Both, listed for the API | Both, covering the API | Not stated for this API |
| HIPAA BAA | Not stated | Standard BAA signed in the Console, if eligible | Standard BAA accepted in the API Platform, if eligible | Not stated for this API |
| Data regions | EU and US endpoints at +10% | US-only inference at 1.1x; otherwise global | 10 storage regions by approval; processing in US, EU, UAE | No region choice stated |
| Official SDKs | Python, TypeScript | Python, TypeScript, C#, Go, Java, PHP, Ruby | Python, TypeScript/JavaScript, Java (beta), Go (beta), C#, Ruby | Python, JavaScript/TypeScript, Go, Java, C# |
| Console SSO | SAML on the Enterprise plan | Through sales | Custom/Enterprise billing plan | Not stated (Google account sign-in) |
| Bottom line | From its public pages, the Mistral AI API has the lowest reference price in this comparison, EU or US endpoints you can pick yourself and spending caps. Training use is an opt-out, rate limits sit behind sign-in, and ZDR and SSO need approval or the Enterprise plan. | From its public pages, the Anthropic API has a complete per-token price list, discounts that combine and strong data terms, with a self-serve HIPAA BAA. Region choice is US only, zero data retention and SSO go through sales, and a few pages disagree on billing details. | From its public pages, the OpenAI API has a complete per-token price list, published batch, flex and caching discounts, and audit reports that cover the API. Stricter data controls, residency and SSO go through sales, and a few docs pages lag behind the price list. | From its public pages, the Gemini API pairs an ongoing free tier on current models with a detailed USD price list and published batch, flex and caching discounts. Certifications, zero data retention and data regions sit with Google Cloud's Agent Platform, and a few billing pages disagree. |
Evidence: Mistral AI API: facts checked 2026-10-15 from its public sign-up and pricing pages, no completed hands-on run; Anthropic API (Claude): facts checked 2026-10-15 from its public sign-up and pricing pages, no completed hands-on run; OpenAI API: facts checked 2026-10-15 from its public sign-up and pricing pages, no completed hands-on run; Google Gemini API: facts checked 2026-10-15 from its public sign-up and pricing pages, no completed hands-on run. Prices in USD as listed, excluding tax. "–" means not recorded.
Score
- Pricing clarity
- 9.0 / 10, Excellent
- per-token prices published for every model; paid online; Mistral Large 4 sale price has no end date
- Free tier
- 6.0 / 10, Good
- free tier that does not expire; limits not published; free-tier prompts may be used to improve models
- Price for 100M input + 20M output tokens
- 8.1 / 10, Excellent
- $12/mo on ministral-3b-2512: 100M input tokens at $0.10 and 20M output tokens at $0.10 per million (computed from the USD pricing page, global endpoint, standard tier)
- Data use and retention
- 6.5 / 10, Good
- used for training unless you opt out; kept 30 days by default; zero data retention by approval; DPA with the terms
- Compliance and regions
- 8.0 / 10, Excellent
- own SOC 2 Type II, ISO 27001, regional processing: EU, US
- Cost controls and developer tools
- 6.8 / 10, Good
- batch 50% off, cached input 90% off, rate limits not published, SDKs: Python, TypeScript, structured output, compatible endpoint, spend limits, SSO by sales
Review
Review from public pages
Written from Mistral AI API's public pages (pricing, security or privacy, and docs) as read on Oct 15, 2026. We did not sign up, and no hands-on run is published yet.
What the public pages show
We read Mistral AI's API pricing page, the Ministral 3 3B and other model pages, the docs on billing, subscriptions, spending limits, batch processing, prompt caching, regional inference, Priority Tier, zero data retention, SAML SSO, SDKs and known limitations, the help center articles on training, ZDR, rate limits, certifications and data location, and the legal center (Commercial Terms, DPA, Additional Terms, Privacy Policy) on Oct 15, 2026; we did not sign up, create an API key or call the API. The docs pricing table shows USD by default, and every price here is the USD figure. The trust center only loads with JavaScript, so the certifications come from the help center.
The money side is clear. Every new account starts in Free mode with API keys and no card; paid use is prepaid pay-as-you-go credits at list price, with optional auto recharge and monthly spending limits per Organization and per Workspace. The pricing page lists every current text model per 1M tokens with Batch (50% off), Priority (1.75x) and Regional (+10%) toggles, and the caching docs bill cached prompt tokens at 10% of the input price. Ministral 3 3B is GA, has a 256k context and supports function calling, so it sets the reference price.
A few points need a question. The Free mode allowance and every rate-limit number appear only on the Admin Limits page after sign-in. Training use is an opt-out on Free mode and pay-as-you-go alike, and the Commercial Terms say opt-outs, including ZDR, do not apply to Labs or Preview models, meaning experimental, preview or pre-release models identified as such. Mistral Large 4 is listed as Public Preview, while the ZDR docs exclude only Labs models. Some docs pages disagree: the batch guide allows up to 1 million requests per batch while the known-limitations page says 100,000, and rate limits are described per organization on one page and per Workspace on another.
Strengths
- The lowest reference price in this comparison
Ministral 3 3B costs $0.10 per 1M input and output tokens, so 100M input and 20M output tokens come to $12 a month at standard rates (computed); Batch would halve it.
vs OpenAI API: The same tokens cost $20 on OpenAI API (GPT-6 Luna) and on Anthropic API (Claude Haiku 5.5), computed from their pricing pages.
- EU or US endpoints you choose yourself
The api.eu.mistral.ai and api.us.mistral.ai endpoints keep inference in EU and EFTA or in US data centers at 1.1x list price, with no sales step stated.
vs OpenAI API: Anthropic API (Claude) offers only a US option; OpenAI API data residency goes through sales.
- Free mode with API keys and no card
The quickstart says API access is enabled by default in Free mode with no credit card required, with usage and rate limits applied.
vs DeepSeek API: OpenAI API states no free allowance, and DeepSeek API states no free tier.
- Spending caps per Organization and Workspace
The billing docs describe a monthly spending limit for the Organization and separate limits per Workspace, plus an email to Admin and Billing members when 80% of the last top-up is used.
vs DeepSeek API: DeepSeek API states no spend limit; spending is bounded only by the prepaid balance.
- Lifecycle notice and open weights
The lifecycle policy gives GA models 6 months of deprecation notice, and the model pages list Ministral 3 and Small 4 weights under Apache 2.0; the pricing FAQ says commercial deployments need a Mistral license, so check the licence before self-hosting.
Things to know
- Training use is an opt-out
The help center says Free mode inputs and outputs may be used for training and that pay-as-you-go customers can opt out at any time. Commercial Terms 4.3 add that opt-outs, including ZDR, do not apply to Labs or Preview models.
Workaround: Turn off "Anonymous improvement data" in Admin › Privacy (the Vibe toggle is separate), and use GA models for sensitive data.
- Free allowance and rate limits are behind sign-in
Free mode limits and all rate-limit numbers appear only on the Admin Limits page. The help center lists tiers by total billed amount ($20, $100, $500, $2,000) and says adding credits does not raise limits.
Workaround: Above Tier 4, the help center says to contact support with the target requests per second, model and token estimates.
- Stricter controls need approval or Enterprise
ZDR is by request and approval on pay-as-you-go, for stateless endpoints only. SAML SSO (no OIDC), audit logs and the Admin API need the Enterprise plan, and no page we read mentions a HIPAA BAA.
Workaround: Request ZDR with the reason for it; without ZDR, the privacy policy keeps API inputs and outputs for 30 rolling days.
- Some docs pages disagree
The batch guide allows 1 million requests per batch, the known-limitations page 100,000; rate limits are per organization on one page and per Workspace on another; the default data location differs; and model pages say Apache 2.0 while the pricing FAQ says commercial use needs a Mistral license.
Workaround: Plan batches of 100,000 requests or fewer, and check the Limits page in the Admin panel after sign-up.
Pricing reality
Mistral bills per token for each model from prepaid credits, after a Free mode whose allowance the public pages do not state. Our scenario on Ministral 3 3B is 100 × $0.10 + 20 × $0.10 = $12 a month at standard rates (computed); Batch halves it, and cached input costs $0.01 per 1M. The same tokens cost $27 on Mistral Small 4 and $300 on Mistral Medium 3.5 (our arithmetic). Costs rise with the EU or US regional endpoints (1.1x list price), Priority Tier (1.75x, set up with Mistral) and Mistral Large 4 once its temporary sale price ends ($1.36 / $4.18 original, no end date stated). Rate-limit tiers rise with total billed amount ($20, $100, $500, $2,000), not with credits added. Plan prices exclude taxes, and the billing docs say credits are deducted including tax.
Exit path
Plain completions are stateless, so there is little to move; uploaded files are kept 30 days unless deleted, and batch results can be downloaded for 24 hours. For customers located in the EU, the Additional Terms (EU Data Act) set out a written switching request sent at least two months ahead, a transitional period of at least 30 days and export of input and output data with its metadata. The OpenAI-style request shape keeps client code portable.
Compared with
- Pick Mistral AI API when
- You want a lower scenario price ($12 on Ministral 3 3B against $20 on GPT-6 Luna), a free mode with no card, and EU or US endpoints without a sales call.
- Pick OpenAI API when
- You want API data kept out of training by default, published rate limits, official SDKs in six languages and a HIPAA BAA path.
- Pick Mistral AI API when
- You want a lower scenario price ($12 against $55), SOC 2 Type II and ISO 27001 statements, ZDR by approval and a choice of EU or US processing.
- Pick Google Gemini API when
- You want paid prompts kept out of training by default, a flex rate at 50% off (preview), and official SDKs in five languages.
- Pick Mistral AI API when
- You need a batch discount, spending limits, a DPA with the terms and EU or US processing; DeepSeek says data is stored in the PRC.
- Pick DeepSeek API when
- You want a 1M-token context on the low-cost model and off-peak rates at half price.
Choose it if
- You want low per-token prices on small models with function calling and JSON-schema output.
- You want processing kept in the EU or the US through a self-serve endpoint.
- You want to try the API in Free mode before adding a card.
- You may later want to run the same open-weight model yourself, after checking its licence for commercial use.
Look elsewhere if
- You need API data kept out of training without changing a setting.
- You need a HIPAA BAA, or official SDKs beyond Python and TypeScript.
- You need published rate limits before you sign up.
Bottom line
From its public pages, the Mistral AI API has the lowest reference price in this comparison, EU or US endpoints you can pick yourself and spending caps. Training use is an opt-out, rate limits sit behind sign-in, and ZDR and SSO need approval or the Enterprise plan.
Plan limits at a glance
| Limit | As recorded |
|---|---|
| How you pay | Prepaid pay-as-you-go credits, optional auto recharge |
| Free option | Free mode, no card; allowance not stated publicly |
| Reference scenario (100M in + 20M out tokens/mo) | $12 on ministral-3b-2512, computed |
| Batch, cache and other discounts | Batch 50% off; cached input 90% off |
| Trained on API data | Unless you opt out (Free mode and paid) |
| Default retention | 30 rolling days for abuse monitoring |
| Zero data retention | By approval, pay-as-you-go, stateless endpoints only |
| SOC 2 / ISO 27001 | Both, stated in the help center |
| HIPAA BAA | Not stated |
| Data regions | EU and US endpoints at +10% |
| Official SDKs | Python, TypeScript |
| Console SSO | SAML on the Enterprise plan |
Read from the vendor's public pages on Oct 15, 2026. No hands-on run is published yet. How we check
We have not published a sign-up run for Mistral AI API yet. The access facts above were read from its sign-up and pricing pages on Oct 15, 2026.
We have not published API calls for Mistral AI API yet.
Mistral AI API lists 6 plans. The cheapest paid plan, Ministral 3 3B, costs $0.10/1M input tokens.
- Mistral AI APIFreeFree mode$0/moDefault state of every new account; no card; allowance shown only after sign-in
- API keys and Studio
- Limited usage and rate limits
- Inputs and outputs may be used for training unless you opt out
- Mistral AI APIMinistral 3 3B$0.10/1M input tokens$0.10 per 1M output tokens; cached input $0.01; paid from prepaid credits
- 256k context window
- Function calling and structured outputs
- Batch at 50% off
- Apache 2.0 open weights
- Mistral AI APIMistral Small 4$0.15/1M input tokens$0.60 per 1M output tokens; cached input $0.015
- Batch at 50% off
- Regional price +10% on the pricing page (models by region not stated)
- Mistral AI APIMistral Large 3$0.50/1M input tokens$1.50 per 1M output tokens; cached input $0.05
- Batch at 50% off
- Priority price 1.75x on the pricing page (eligibility set up with Mistral)
- Mistral AI APILowest pricePro$14.99/moExcluding taxes; one plan across Vibe, Studio and API usage
- $25.5 /mo in API credits
- Chat and email support
- Usage beyond the plan at API rates
- Mistral AI APIEnterpriseContact salesContact us on the plans page
- SAML SSO
- Audit logs
- Admin API
- Custom SLAs and dedicated support
Prices as listed, USD, excluding tax; observed 2026-10-15.
Source: docs.mistral.ai pricing page. Prices change; the observed date above is when our agent read them.