Claude Haiku 4.5 vs Codestral

Codestral is cheaper for every workload modelled here.

List prices checked

Specifications and rates

PropertyClaude Haiku 4.5Codestral
Input per 1M$1.00$0.300
Output per 1M$5.00$0.900
Cached input per 1M$0.100
Blended (3:1)$2.00$0.450
Context window200K262K
Max output64K
Inputs acceptedtext, imagetext

Source: Anthropic pricing (August 2026) · Mistral AI pricing (May 2026). Published list prices only — committed-use discounts, regional variation, and per-request charges are not included. Not yet independently re-checked against the vendor’s page.

Monthly cost by workload

Identical token volumes through both rate cards. Output-heavy profiles are where the ranking flips.

WorkloadClaude Haiku 4.5CodestralCheaper
In-app chat assistantA support or help assistant handling a few thousand conversations a month, with short prompts and short answers.$45$11Codestralsaves $35/mo (77%)
RAG over documentsRetrieval-augmented answering where each request stuffs several retrieved passages into the prompt, so input dominates.$375$104Codestralsaves $272/mo (72%)
Coding agentAn agentic loop that reads files and writes patches, generating heavy output and re-sending large context on every turn.$900$210Codestralsaves $690/mo (77%)

What each one is for

Claude Haiku 4.5

by Anthropic

The fast, cheap end of the Claude family, priced at a third of Sonnet while keeping the same context window. Suited to classification, extraction, and high-volume routing.

Codestral

by Mistral AI

A code-specialised model tuned for fill-in-the-middle completion, priced for the high request volume that inline editor autocomplete generates.

Common questions

Is Claude Haiku 4.5 or Codestral cheaper in 2026?
Codestral, for every workload modelled here. On a blended 3:1 basis it costs $0.450 per million tokens against Claude Haiku 4.5's $2.00.
Which has the larger context window?
Codestral, at 262K tokens against 200K. A larger window only helps if you can afford to fill it — the cost of one full-context request is in the table above.
Which is cheaper for output-heavy work like agents?
Codestral, at $0.900 per million output tokens against $5.00. Reasoning tokens bill as output, so for agentic loops this axis usually decides the invoice regardless of what the input rates look like.
What is not included in these figures?
Batch-processing discounts, fine-tuning, long-context surcharges above a provider's standard tier, and any enterprise agreement. These are published pay-as-you-go list prices, checked against each provider's own pricing page on the date shown.
Is switching from Claude Haiku 4.5 to Codestral worth it?
On price alone it is about 78% cheaper on a blended basis. Whether that survives contact with your workload depends on output quality and on retry rate — a cheaper model that needs two attempts is not cheaper. Model the specific volumes in the calculator before you migrate.

Other model comparisons