The short answer
Codestral 2508 is a Mistral model priced at $0.3 per million input tokens and $0.9 per million output tokens, with a 256K token context window. One standard answer, meaning 1,000 tokens in and 500 tokens out, costs $0.0008. Inside Whizi it costs 1 credit per message and needs the Powerhouse plan or above.
That makes it one of the cheaper models in the catalogue: 63 of the 89 priced models cost more per answer, and 25 cost less.
What Codestral 2508 costs
List rates are what the provider charges per token. The cost of one standard answer is the more useful number, because it prices the same unit of work across every model regardless of how each one is billed.
| Input, per million tokens | $0.3 |
| Output, per million tokens | $0.9 |
| One standard answer | $0.0008 |
| One thousand answers | $0.7500 |
| Context window | 256K tokens |
| Credits per message in Whizi | 1 |
| Minimum Whizi plan | Powerhouse |
| Rank by cost, of 89 priced models | 26 |
Prices are the published Mistral rates as carried by OpenRouter, last refreshed 2026-08-08. The full dataset covering all 100 models from 28 providers is the AI model cost index, which is free to reuse with attribution.
What it costs inside Whizi
Whizi meters messages in credits rather than tokens, and Codestral 2508 is charged at 1 credit per message. The cost does not change with the length of your message or the length of the answer, so a one line question and a long document analysis cost the same.
| Plan | Price per month | Monthly credits | Messages on this model |
|---|---|---|---|
| Powerhouse | $49.99, or $34.99 billed yearly | 8,000 | 8,000 |
Codestral 2508 is part of the Powerhouse catalogue. Anything not explicitly listed on the Starter or Pro tiers resolves to Powerhouse, which is the deliberate default for the frontier and specialist models. See plans and limits.
How it compares on price
Against the cheapest model in the index, Ling-2.6-flash at $0.000025 per answer, Codestral 2508 costs 30 times more. Against the median, GLM 4.6 at $0.0015, it costs 2.0 times less.
Its nearest neighbours from Mistral:
| Model | One answer | Context | Credits |
|---|---|---|---|
| Codestral 2508 (this page) | $0.0008 | 256K | 1 |
| Mistral Large 3 2512 | $0.0013 | 262K | 2 |
| Mistral Medium 3.1 | $0.0014 | 131K | 2 |
| Mistral Medium 3.5 | $0.0053 | 262K | 6 |
It is the cheapest Mistral model in the priced index, so within this provider there is nothing to route down to.
The context window in practice
Codestral 2508 accepts 256K tokens of context, which is roughly 380 pages of text. That is below the median for the priced catalogue.
Advertised context and usable context are not the same thing. Recall tends to degrade before the stated limit is reached, particularly for material in the middle of a long input, so the practical test is to upload a long document and ask about something buried halfway through rather than trusting the number.
The number matters most when you are working from your own files. If a document is too long for the model you picked, switching to a larger-context model in the same conversation carries the thread across, so you are not starting over. See switching models mid-conversation.
- Input $0.3 and output $0.9 per million tokens
- $0.0008 for one standard answer, $0.7500 for a thousand
- 256K token context window
- 1 credit per message inside Whizi
- Requires the Powerhouse plan or above
- Ranks 26 of 89 priced models by cost per answer
Frequently asked questions
How much does Codestral 2508 cost?
$0.3 per million input tokens and $0.9 per million output tokens at list rates, which works out to $0.0008 for one standard answer of 1,000 tokens in and 500 out, or $0.7500 for a thousand of them. Inside Whizi it is 1 credit per message on a monthly allowance, rather than metered per token.
Which Whizi plan includes Codestral 2508?
Powerhouse and above, which is $49.99 per month or $34.99 billed yearly. At 1 credits per message, a Powerhouse allowance of 8,000 credits covers 8,000 messages on this model if you spend the whole allowance here, and most people mix in a 1 credit model for routine work so it goes further.
What is the context window of Codestral 2508?
256K tokens, which is roughly 380 pages of text and sits below the median for the priced catalogue. Treat that as the ceiling rather than the working figure: recall usually degrades before the stated limit, especially for content in the middle of a long input.
Is Codestral 2508 expensive compared to other models?
It ranks 26 of 89 priced models by cost per answer, so 63 cost more and 25 cost less. The spread across the whole catalogue is about 4800 times from cheapest to priciest, which is far wider than most people assume, so "expensive" only means anything relative to the specific alternative you would otherwise use.