Mistral's current API model lineup
Mistral prices three models for API access: Mistral Large 3, Mistral Small 4, and Codestral. Mistral Large 3 sits at the top of the lineup as the flagship, priced above the other two on both input and output tokens. Mistral Small 4 is the budget option, priced well below Large 3 and suited to high-volume or simpler work. Codestral is Mistral's code-focused model, priced between the two general-purpose models, a distinct offering rather than a variant of Large or Small, aimed specifically at code generation and completion tasks rather than general chat or reasoning.
The Mistral Large 3 repricing
One detail worth knowing before you compare Mistral against notes from more than a few months back: Mistral Large 3, the "2512" release, cut pricing by roughly 75% compared with the previous Large 2 generation. That's a large enough shift that any cost comparison built on Large 2's rate (whether from an old blog post, a cached spreadsheet, or a prior version of this kind of calculator) will substantially overstate what Mistral's flagship actually costs today. If you're re-evaluating Mistral after previously ruling it out on price, it's worth another look with current numbers.
How Mistral's pricing structure works
Like the rest of the market, all three Mistral models bill input and output tokens separately, with output priced higher across the lineup. What stands out about Mistral specifically is where its rates land relative to the market as a whole: even its flagship, Large 3, is priced well below flagship-tier models from the larger US providers, which is a large part of why cost-sensitive teams evaluate Mistral in the first place. That positioning holds down through Small 4 and Codestral too, making the whole lineup a reasonable default to check whenever budget is the primary constraint on model choice.
When Codestral is the right call
If your workload is genuinely code-focused (generation, completion, refactoring, or code review), Codestral is worth comparing directly against both Small 4 and Large 3 rather than defaulting to whichever general-purpose model you already use elsewhere in your product. A model built and priced specifically for code can outperform a general-purpose model on that narrow task at a lower cost than the general-purpose flagship, which is exactly the kind of task-specific routing that keeps a multi-model setup cost-efficient.
Cutting your Mistral API bill
Route high-volume or straightforward requests to Mistral Small 4 rather than Large 3 by default, and reserve Large 3 for tasks that genuinely benefit from the extra capability. If your traffic includes any meaningful volume of code-related work, route it to Codestral specifically rather than a general-purpose model. Beyond model routing, cap output length so responses don't run longer than needed, and keep system prompts and few-shot examples as lean as they can be while holding quality, the same fundamentals that apply across every provider, but worth combining here with Mistral's already-favorable base rates for a genuinely low-cost setup.
A worked example on Mistral Large 3
At Mistral Large 3's current rate ($0.50 input / $1.50 output per million tokens), a 1,000-token prompt with a 500-token answer costs 1,000 ÷ 1,000,000 × $0.50 = $0.0005 for input, plus 500 ÷ 1,000,000 × $1.50 = $0.00075 for output, $0.00125 per call, or $125 a month at 100,000 calls. Run the same volume on Mistral Small 4 instead and the bill drops further, since Small 4's rate sits well below Large 3's on both input and output, exactly the kind of side-by-side check worth running with your own token counts before choosing a default tier.
Compare Mistral against OpenAI's GPT API pricing, the similarly budget-focused DeepSeek API pricing calculator, or Google's Gemini API pricing. For every provider in one view, use the all-provider hub calculator.
