What affects cost
- Selected Mistral model and service
- Input and output token volume
- Cached versus uncached repeated context
- Synchronous versus Batch processing
- OCR, audio, fine-tuning, or dedicated deployment units
LLM API
Mistral API pricing
Mistral API pricing separates standard input and output tokens from cached-input and Batch processing discounts.
Decision summary: Consider Mistral when its model portfolio, open-weight options, deployment choices, and token discounts fit the workload and operating requirements.
Pricing overview
Usage-based per-model token pricing with documented Batch and cached-input discounts; other products and deployment modes use separate terms.
Source-tracked model data
Mistral API model prices in one table
Example cost uses 1,000,000 input tokens and 300,000 output tokens at each model's default tracked rate, with no cache discount.
| Model | Input price | Cached input | Output price | Context | Example cost |
|---|---|---|---|---|---|
| Mistral Medium 3.5 Stable | $1.50 / 1M | $0.15 / 1M | $7.50 / 1M | 256K tokens | $3.75 |
Cost boundary: this example compares token charges only. It does not assume equal output quality, latency, retry rates, or output length across models.
Lower-cost options from the same provider
- Mistral Small for cost-sensitive tasks after quality testing
- Batch processing for eligible asynchronous jobs
- Cached input for repeated prompts and stable shared context
Alternative providers or products
- OpenAI API
- Anthropic API
- Google Gemini API
- Together AI or GroqCloud for supported open-model inference
Best for
- teams evaluating Mistral-hosted API models
- batch workloads that can use asynchronous processing
- repeated-context workloads eligible for cached-input discounts
Not ideal for
- buyers treating every Mistral product as one token rate
- real-time workloads modeled using Batch discounts
- teams that have not verified the exact active model identifier
FAQ
Mistral API pricing questions
How is Mistral API usage billed?
Text-model usage is generally billed per million input and output tokens, with separate discounts or units for caching, Batch, OCR, audio, fine-tuning, and deployment products.
Can every workload use the 50% Batch discount?
No. Batch is intended for asynchronous processing. Real-time requests should be budgeted using the applicable standard model rate.