API model string
mistral-large-3Context window
128K tokens
Max output not published
- Knowledge cutoff
- not published
- Released
- 2025–2026
- Modalities
- text in, text out
Last verified July 10, 2026
Rate limits by tier
Mistral La Plateforme organizes access into three tiers: a free Experiment tier for evaluation, a pay-as-you-go tier unlocked by billing card, and a custom Enterprise tier negotiated with sales. Exact RPM figures are not published publicly — your live org ceilings appear in Admin Console → Limits. This page covers Mistral Large 3, the current flagship; 'Mistral Large' historically referred to Large 2, which this version supersedes.
| Tier | Requirements | RPM | TPM | RPD | Concurrent | Notes |
|---|---|---|---|---|---|---|
| Free (Experiment) | No billing card required — sign up at mistral.ai and access the API immediately for evaluation | not published (check Admin Console → Limits) | ~1B tokens/month cap | not published | — | Rate-limited and for evaluation only — not suitable for production. The -latest alias (mistral-large-latest) can silently pull a new major version; pin 'mistral-large-3' in production. |
| Pay-as-you-go (paid) | Enable billing in La Plateforme account settings | ~300 RPM baseline (per third-party trackers — confirm in Admin Console) | not published | not published | — | No monthly minimums. Mistral Large is the flagship tier. Limits grow with usage history. Pin 'mistral-large-3' explicitly rather than using the -latest alias. |
| Enterprise (custom) | Contact Mistral sales | not published | not published | not published | — | SAML SSO, audit logs, custom limits, dedicated SLA. Mistralship startup program offers ~$30K credits via application — apply at mistral.ai/mistralship. |
Swipe the table sideways to see every limit column.
- 1.Mistral no longer publishes exact RPM numbers publicly; Admin Console → Limits is the authoritative source for your organization's current ceilings.
- 2.Pricing is inconsistent across trackers: Mistral Large 3 is cited at $2.00/$6.00 (widely-used figure, used here) AND at $0.50/$1.50 on other trackers. Verify on mistral.ai/pricing before quoting.
- 3.The $6/MTok output figure is marketed by Mistral as the cheapest flagship-tier output among Western providers as of July 2026 — verify against competitors before using this claim.
Limits verified against the Mistral AI docs, July 10, 2026.
Token pricing
What you pay per million tokens (USD). Input and output are billed separately.
Input
$2.00
per 1M tokens
Output
$6.00
per 1M tokens
Batch discount
−50%
on batch jobs
- The $2.00/$6.00 per MTok figures are the most widely-cited values for Mistral Large 3 from third-party trackers and are used here. Some trackers list Large 3 at $0.50 in / $1.50 out — the discrepancy is unresolved. Verify on mistral.ai/pricing before populating any budget calculation.
- Context caching is available and discounts repeated-prefix input; exact cached rate not published — verify on mistral.ai/pricing.
- Batch API provides 50% off input and output for async, non-latency-sensitive workloads.
- Reasoning/thinking tokens (applicable to Magistral models only) are billed at the output rate — this does not directly apply to Mistral Large 3.
Side-project assistant
$16.00
per month
Assumptions
5M tokens in / 1M tokens out per month
5M × $2.00/MTok + 1M × $6.00/MTok = $10.00 + $6.00 = $16.00 (verify rate on mistral.ai/pricing)
Growing SaaS
$108
per month
Assumptions
60M tokens in / 12M tokens out per month (no published cached rate — floor without caching)
60M × $2.00/MTok + 12M × $6.00/MTok = $120 + $72 = $192 floor; with batch discount on half: $96 + $36 = $132 on batched portion; blended ~$108 (estimate — cached rate TBD)
Enterprise pipeline
$1,280
per month
Assumptions
400M tokens in / 80M tokens out per month
400M × $2.00/MTok + 80M × $6.00/MTok = $800 + $480 = $1,280 (verify rate; batch at 50% off = $640)
Run your own numbers
Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.
Estimated mistral-large-3 spend
$90/mo
Input: $60
Output: $30
30M in × $2.00 + 5M out × $6.00 = $90
Same volume, priced across models
- Mistral Medium 3.5Cheapest$22
- Devstral$55
- Gemini 2.5 Pro$88
- mistral-large-3This model$90
Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to mistral-large-3 only. Estimates for comparison; real bills vary with request shape and long-context surcharges.
mistral-large-3 vs the alternatives
Mistral Large 3 positions as the cost-efficient Western flagship — cheaper output than GPT and Gemini Pro, with EU data residency by default.
| Aspect | mistral-large-3 | Gemini 2.5 Pro | Mistral Medium 3.5 | Devstral |
|---|---|---|---|---|
| Context window | 128K | 1M+ | 128K | 200K |
| Input $/MTok | $2.00 (verify) | $1.25 (≤200K) | $0.40–1.50 (verify) | $1.00 |
| Output $/MTok | $6.00 (verify) | $10.00 (≤200K) | $2.00–7.50 (verify) | $5.00 |
| EU data residency | Yes (Paris) | No | Yes (Paris) | No |
| Long-context surcharge | No (flat to 128K) | Yes (>200K: 2×/1.5× price) | No | No |
| Batch discount | 50% | 50% | 50% | 50% |
| Free eval tier | Yes (Experiment) | No (paid-only since Apr 2026) | Yes (Experiment) | No |
| Cheapest output among flagships | Yes ($6 vs $10–$30 at verify) | No ($10) | — | — |
Swipe the table sideways to see every model.
Hitting a 429? The playbook
The exact errors you'll see
HTTP 429 Too Many Requestsrate_limit_errorRetry-After: <seconds>Why it happens & how to fix it
Experiment tier RPM ceiling hit
Switch to pay-as-you-go — add a billing card in La Plateforme account settings. The free tier is designed for evaluation, not production throughput.
Paid baseline ~300 RPM exceeded during burst
Implement Retry-After backoff and space requests. Check Admin Console → Limits for self-serve increase options. Use the Batch API for non-latency-sensitive workloads to avoid RPM pressure.
-latest alias silently upgraded to a new Large version
Pin 'mistral-large-3' explicitly in all production model strings. The -latest alias can silently switch to a future Large version with different pricing and behavior.
Monthly ~1B token cap on free tier exhausted
Add a billing card to upgrade to pay-as-you-go; the paid tier has no monthly token cap.
Context caching misconfiguration causing higher-than-expected token burn
Verify cache hits in the usage dashboard. Ensure the shared prefix is stable between requests — any change to the prefix invalidates the cache. Restructure prompts to maximize prefix reuse.
Retry strategy
HTTP 429 returns a Retry-After header specifying wait seconds — honor it precisely. Use exponential backoff as a fallback: delay = min(2^attempt, 60) seconds ± 10% jitter. La Plateforme is OpenAI-compatible; initialize the OpenAI SDK with base_url='https://api.mistral.ai/v1' and substitute your Mistral API key.
1import OpenAI from 'openai';23const client = new OpenAI({4 apiKey: process.env.MISTRAL_API_KEY,5 baseURL: 'https://api.mistral.ai/v1',6});78async function callMistralLargeWithRetry(9 prompt: string,10 maxRetries = 511): Promise<string> {12 for (let attempt = 0; attempt < maxRetries; attempt++) {13 try {14 const response = await client.chat.completions.create({15 model: 'mistral-large-3', // pin explicit version, not mistral-large-latest16 messages: [{ role: 'user', content: prompt }],17 });18 return response.choices[0].message.content ?? '';19 } catch (err: unknown) {20 const error = err as { status?: number; headers?: Record<string, string>; message?: string };21 if (error.status === 429) {22 const retryAfter = parseFloat(error.headers?.['retry-after'] ?? '0');23 const jitter = 1 + (Math.random() * 0.2 - 0.1);24 const delay = retryAfter > 025 ? retryAfter * 100026 : Math.min(Math.pow(2, attempt) * 1000 * jitter, 60000);27 console.warn(`Rate limited. Retrying in ${(delay / 1000).toFixed(1)}s (attempt ${attempt + 1}/${maxRetries})`);28 await new Promise(resolve => setTimeout(resolve, delay));29 } else {30 throw err;31 }32 }33 }34 throw new Error('Max retries exceeded for Mistral Large API call');35}How to raise your limits
The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.
Experiment (free, no card)
ImmediateSign up at mistral.ai — no card required. Access Mistral Large 3 immediately for testing and evaluation.
Unlocks: ~1B tokens/month cap, low RPM (check Admin Console → Limits), eval only
Pay-as-you-go
Immediate upon card additionAdd a billing card in La Plateforme account settings. No minimum monthly spend.
Unlocks: ~300 RPM baseline (per third-party trackers — verify in Admin Console), no monthly token cap, production access
Admin Console → Limits (self-serve)
Immediate for self-serve increasesNavigate to Admin Console → Limits to see your current org ceilings. Some increases are available without contacting support.
Unlocks: Higher org-level limits without a sales call
Mistralship (startup credits)
Application review processApply at mistral.ai/mistralship. Application-only process for ~$30K in credits for qualifying startups.
Unlocks: ~$30K in La Plateforme credits to offset compute costs during growth
Enterprise
Sales negotiationContact sales@mistral.ai for custom terms.
Unlocks: Custom RPM/TPM limits, SAML SSO, audit logs, dedicated SLA, higher rate ceilings
Cut your token spend
Leverage cheapest flagship output pricing
$6/MTok output undercuts GPT-5.x ($15–$30) and Gemini 2.5 Pro ($10) significantly (verify all figures)For output-heavy workloads (long-form generation, detailed analysis, structured JSON output), route to Mistral Large 3 over comparable Western flagship models. The output price differential is the primary cost lever — verify on mistral.ai/pricing before committing.
Context caching
Discount on repeated-prefix input (exact rate not published — verify on mistral.ai/pricing)Structure prompts with a long, stable shared system-prompt prefix — product documentation, persona, tool definitions. Mistral caches repeated prefixes and discounts subsequent input tokens that hit the cache.
Batch API
-50% on input and output cost for async workloadsSubmit document processing, classification, or offline summarization tasks as batches. Use the paid baseline RPM for latency-sensitive requests and the Batch API for everything else to maximize throughput at lower cost.
Pin model version
Prevents silent price changes as new Large versions releaseUse 'mistral-large-3' not 'mistral-large-latest' in all production deployments. The -latest alias can silently pull a new major version with different pricing or output characteristics.
EU residency at no premium
GDPR-adjacent compliance with zero additional cost or configurationMistral routes all La Plateforme traffic through Paris by default. No regional surcharge, no special configuration — unlike AWS/Azure residency options that typically add cost.
Downgrade to Mistral Medium for simpler tasks
Up to ~80% cost reduction on tasks that don't require flagship qualityRoute classification, simple Q&A, and low-stakes summarization to Mistral Medium 3.5. Reserve Mistral Large 3 for complex reasoning, multi-step analysis, and tasks where quality is the primary constraint.
Frequently asked questions
How do I increase Mistral Large rate limits?
Add a billing card to your La Plateforme account to immediately upgrade from the free Experiment tier to the pay-as-you-go baseline (~300 RPM per third-party trackers). Then check Admin Console → Limits for your current org ceilings — some self-serve increases are available. For custom enterprise limits, contact Mistral sales or apply to the Mistralship program at mistral.ai/mistralship.
Is Mistral Large API free?
Yes — Mistral's free Experiment tier requires no billing card and includes approximately 1 billion tokens per month with a low RPM ceiling (check Admin Console → Limits for your specific ceiling). It is for evaluation only. Paid pay-as-you-go access has no monthly minimum and unlocks the production-grade baseline.
What is Mistral Large 3 pricing?
The most widely-cited figures are $2.00 per million input tokens and $6.00 per million output tokens — though some third-party trackers list $0.50/$1.50. Verify on mistral.ai/pricing before budgeting. A 50% batch discount is available for async workloads. Context caching discounts repeated-prefix input at a rate not yet published.
Mistral Large vs GPT-4 / GPT-5 — which is cheaper?
At the $6.00/MTok output figure (verify on mistral.ai/pricing), Mistral Large 3 is marketed as the cheapest flagship-tier output among Western providers — undercutting GPT-5.x at $15–$30/MTok and Gemini 2.5 Pro at $10/MTok. If the $0.50/$1.50 figure from some trackers is accurate, the advantage is even larger. Mistral also includes EU Paris data residency at no additional charge. Verify all pricing before making a final decision.
Does Mistral Large support EU data residency?
Yes. All La Plateforme traffic routes through Paris, France by default. This provides EU data residency without additional cost or configuration — no regional surcharge, unlike AWS/Azure residency options. This applies to Mistral Large 3 and all other La Plateforme models.
What is the difference between Mistral Large and Mistral Large 3?
Mistral Large historically referred to Large 2; the current version is Mistral Large 3. This is a major version upgrade within the same product line. Use the model string 'mistral-large-3' (or 'mistral-large-latest' with caution) in your API calls. Pricing has changed between versions — verify the current rate on mistral.ai/pricing, as Large 2 and Large 3 rates differ across third-party trackers.
How do I fix Mistral Large 429 rate limit errors?
Read the Retry-After header in the 429 response and wait that many seconds before retrying. Use exponential backoff as a fallback: delay = min(2^attempt, 60) seconds ± 10% jitter. If on the free tier, add billing to access the paid ~300 RPM baseline. For burst traffic, use the Batch API for non-latency-sensitive work, and check Admin Console → Limits for self-serve increase options. RapidDev helps teams architect retry strategies for Mistral workloads — rapidevelopers.com/contact.
Should I use mistral-large-latest or pin the version?
Pin the explicit version string 'mistral-large-3' in all production deployments. The -latest alias can silently pull a new major version mid-deploy with different pricing, behavior, and potentially different rate limits. Using -latest is fine for rapid prototyping, but any production system should pin an explicit version and update intentionally after testing.
We build AI apps that don't hit rate limits
- Retry, backoff & caching built in
- Multi-provider fallback routing
- Fixed price, you own the code
30-min call. No commitment.