Mistral API Pricing 2026:
Mistral Medium 3.5, Large 3, Small 4 & Current Models
Current Mistral AI API pricing for Medium 3.5, Large 3 and Small 4, plus retired models, EU deployment advantages and a comparison with OpenAI, Google and Claude. Last verified: 2026-10-01.
Current Mistral Models
| Model | Input / 1M | Output / 1M | Context | Licence · Best for |
|---|---|---|---|---|
| Mistral Medium 3.5 | $1.50 | $7.50 | 256K tokens | Modified MIT · Mistral's frontier-class multimodal model for agents and coding |
| Mistral Large 3 | $0.50 | $1.50 | 256K tokens | Apache 2.0 · Open-weight general-purpose multimodal model |
| Mistral Small 4 | $0.15 | $0.60 | 256K tokens | Apache 2.0 · Instruct, reasoning and coding in one efficient model |
Medium 3.5 is newer and more capable than Large 3 despite the name — Mistral describes it as its frontier-class model, while Large 3 remains the cheaper open-weight generalist.
Why Choose Mistral?
Mistral stands out in the 2026 AI API landscape for four reasons:
- EU-friendly data residency: Mistral operates under EU law and offers EU-hosted deployment, making it attractive for GDPR-sensitive workloads
- Open-weight models: Large 3 and Small 4 are Apache 2.0 and Medium 3.5 uses a modified MIT licence, so you can self-host them
- Competitive pricing: Mistral Large 3 at $0.50/M input costs a quarter of GPT-6.1 Sol or Claude Sonnet 5.5 ($2.00/M); Medium 3.5 at $1.50/$7.50 also undercuts both
- Commercial-friendly licensing: the open-weight models can be used in commercial products
Mistral vs Other Providers — Production Comparison
| Model | Provider | Input / 1M | Output / 1M | Context |
|---|---|---|---|---|
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | 1M | |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | 1.05M |
| Mistral Small 4 | Mistral | $0.15 | $0.60 | 256K |
| Mistral Large 3 | Mistral | $0.50 | $1.50 | 256K |
| Gemini 3.8 Flash | $0.75* | $3.75* | 1M | |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | 200K |
| Mistral Medium 3.5 | Mistral | $1.50 | $7.50 | 256K |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | 1M |
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | 1.05M |
*Promotional rate through December 31, 2026; $1.50/$7.50 from January 1, 2027.
Mistral Small 4 is no longer the cheapest model on the market — Gemini 2.5 Flash-Lite and GPT-6 Luna both start at $0.10/M input — but it is the cheapest open-weight option from a major provider. Mistral models also tend to do well on multilingual and European-language workloads.
Codestral — For Code-Specific Workloads
Codestral is Mistral's specialized code generation model. It supports fill-in-the-middle (FIM) prompts — useful for IDE integrations and code completion systems. Check current pricing in Mistral's official docs before publishing production cost estimates, as specialized model pricing can change independently of text model pricing.
Retired and Legacy Models
| Model | Status | Replaced by |
|---|---|---|
| Mistral Small 3.2 | Deprecated 2026-04-30, retired 2026-07-31 | Mistral Small 4 |
| Mistral Medium 3.1 | Deprecated 2026-05-22, retired 2026-08-31 | Mistral Medium 3.5 |
| Mistral Small 3.1 | Retired | Mistral Small 4 |
| Mistral Large 2 | Older generation | Mistral Large 3 |
mistral-small-2506 should move to Mistral Small 4 ($0.15/$0.60) — slightly more expensive per token, with a 256K context window instead of 128K. Self-Hosting Mistral Models
A key differentiator from OpenAI and Anthropic: Mistral publishes model weights for self-hosted deployment. This means:
- Inference cost at scale can be dramatically lower than API pricing
- Full control over data — no third-party data processing
- On-premise or private cloud deployment for EU compliance requirements
- Fine-tuning on proprietary data without sharing it with Mistral
Self-hosting requires GPU infrastructure. For small and medium workloads, the API is more economical; for very large deployments, self-hosting can reduce costs substantially.
Mistral API Access
- Access via console.mistral.ai
- Pay-as-you-go — no monthly minimum
- EU-hosted option available for data residency requirements
Frequently Asked Questions
Which Mistral model should I use in 2026?
Mistral Small 4 for high-volume, cost-sensitive work; Mistral Large 3 when you want a capable open-weight generalist at $0.50/M input; and Mistral Medium 3.5 for the strongest results on agentic and coding tasks.
What replaced Mistral Small 3.2?
Mistral Small 4. Small 3.2 was deprecated on 2026-04-30 and retired on 2026-07-31. Small 4 costs $0.15/M input and $0.60/M output with a 256K context window.
Is Mistral Medium 3.5 better than Mistral Large 3?
Mistral positions Medium 3.5 as its frontier-class model for agents and coding, so for demanding tasks it is the stronger choice. Large 3 is three times cheaper per input token and Apache 2.0-licensed, which makes it attractive for self-hosting and general workloads.
Can I use Mistral for EU GDPR compliance?
Yes. Mistral is a French company operating under EU law and offers EU-hosted inference endpoints. This makes it attractive for organizations with strict data residency requirements.
Does Mistral offer batch pricing?
Mistral offers a Batch API for asynchronous jobs. Check the current discount in Mistral's docs before planning batch workloads.
Compare Mistral vs OpenAI vs Anthropic Costs
Enter your usage volume to see exact monthly costs across all major providers.
Open API Cost Calculator