Mistral AI — Complete API Errors & Reference
Every Mistral La Plateforme error, every model, every EU-specific quirk — with the fix.
Current models & pricing (September 2026)
| Model | API ID | Input / M tok | Output / M tok | Context | Notes |
|---|---|---|---|---|---|
| Mistral Large 3 | mistral-large-2026 | $4.00 | $12.00 | 256K tokens | Flagship model. Strong reasoning, coding, structured output. Competitive with Sonnet 4.6 on many benchmarks. |
| Mistral Medium 3 | mistral-medium-2026 | $1.50 | $4.50 | 128K tokens | Mid-tier. Best price/performance for most workloads. Recommended default. |
| Mistral Small 3.1 | mistral-small-latest | $0.40 | $1.20 | 128K tokens | Fast and cheap. Best for classification, extraction, high-volume routing. |
| Codestral 2 | codestral-latest | $0.90 | $2.70 | 256K tokens | Code-specialized. Fill-in-the-middle, autocomplete, code generation. Optimized for IDE integrations. |
| Pixtral Large | pixtral-large-latest | $4.00 | $12.00 | 128K tokens | Multimodal (text + images). Vision understanding, document analysis, chart parsing. |
| Mistral Embed | mistral-embed | $0.10 | n/a | 8K tokens | Embedding model. 1024-dim vectors. Multilingual with strong French/European language performance. |
Rate limit tiers
| Tier | Requirement | Limits |
|---|---|---|
| Free Tier | New account, no billing | 1 RPS · 500K TPM · rate-limited experimentation |
| Standard | Billing enabled | 5-10 RPS · 2M TPM · scales with usage |
| Enterprise | Contact sales | Custom RPS · dedicated capacity · EU data residency guarantees · SLA |
Documented errors
Related pillar guides
Related blog posts
Frequently asked questions
When should I choose Mistral over OpenAI or Anthropic?
Three main cases: (1) EU data residency requirements — Mistral hosts in Europe under GDPR strictly, (2) French and European language workloads where Mistral's training data advantage shows, (3) cost-sensitive workloads where Medium 3 or Small 3.1 significantly undercuts equivalent tiers from American providers. For general English-language reasoning, Anthropic and OpenAI are still typically stronger.
What model should I use for a new Mistral integration?
Mistral Medium 3 for most workloads — strong performance at $1.50/$4.50, which is significantly cheaper than Sonnet/GPT-5.6 for comparable results. Mistral Large 3 for complex reasoning where quality justifies 3x cost. Small 3.1 for high-volume simple tasks. Codestral 2 for code-specific workloads.
Does Mistral offer EU data residency guarantees?
Yes — La Plateforme's enterprise tier includes contractual guarantees that data is processed and stored within EU. This is a major differentiator vs American providers, especially for German, French, and healthcare/finance workloads bound by strict GDPR interpretation. Standard tier has technical residency but not contractual guarantees; enterprise is what regulated industries need.
What's the difference between La Plateforme and Le Chat?
La Plateforme is the developer API (equivalent to Anthropic Console or OpenAI Platform). Le Chat is the consumer product (equivalent to Claude.ai or ChatGPT). This hub covers La Plateforme's API. Le Chat is a separate product with its own SDK and integration patterns.
How does Mistral's function calling compare to OpenAI's?
Very similar — Mistral's function calling API is deliberately OpenAI-compatible in shape (tools array, tool_calls response, tool_choice control). Migration from OpenAI to Mistral usually only requires changing the base URL and model ID. Schemas transfer directly.
Can I self-host Mistral models?
Yes for the open-weight versions (Mistral 7B, Mixtral 8x7B, historical releases). No for the flagship 2026 models (Large 3, Medium 3, Codestral 2) which are API-only. If you need self-hosted flagship performance, evaluate the open-weight tier against your requirements.
How do Mistral's rate limits work?
Workspace-based — each API key is scoped to a workspace, and rate limits apply per workspace. Standard tier is generous (5-10 RPS, 2M TPM). Enterprise scales with contract. Multiple keys within a workspace share the limit; multiple workspaces don't.
Does Mistral support prompt caching?
Yes — implicit caching kicks in for prompts over specific token thresholds. Cached input tokens bill at reduced rates. Also supports batch API for asynchronous processing at 50% discount. See docs.mistral.ai for the current cache and batch semantics.