Mistral's naming is misleading on price: Medium 3.5 is the more expensive of the two, not the cheaper one. Both have 256K-token context windows, so the decision comes down to what Medium 3.5's positioning buys you. Note that Mistral has deprecated both models: they retire on April 30, 2027, with Mistral Large 4 as the named replacement.
Mistral Large 3 is 3× cheaper on input and 5× cheaper on output ($0.50/$1.50 vs $1.50/$7.50 per 1M). Pay for Medium 3.5 only if its multimodal, agentic and coding focus measurably helps your task — and since both were deprecated on Oct 31, 2026, evaluate Mistral Large 4 for anything new.
| Mistral Medium 3.5 | Mistral Large 3 | |
|---|---|---|
| Provider | Mistral | Mistral |
| Tier | Mid-tier | Mid-tier |
| Input / 1M tokens | $1.50 | $0.50 |
| Output / 1M tokens | $7.50 | $1.50 |
| Cache hit / 1M | — | — |
| Cache write / 1M | — | — |
| Context window | 256K | 256K |
| Notes | Frontier-class multimodal; agentic & coding focus; deprecated Oct 31, 2026, retiring Apr 30, 2027 — replaced by Mistral Large 4 | Former Mistral flagship; deprecated Oct 31, 2026, retiring Apr 30, 2027 — replaced by Mistral Large 4 |
Mistral Large 3 is 3.0× cheaper on input tokens ($0.50 vs $1.50 per 1M). Mistral Large 3 is 5.0× cheaper on output tokens ($1.50 vs $7.50 per 1M).
| Workload | Mistral Medium 3.5 | Mistral Large 3 | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $60.00 | $16.00 | Mistral Large 3 saves $44.00 |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $51.00 | $16.20 | Mistral Large 3 saves $34.80 |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $7.13 | $2.23 | Mistral Large 3 saves $4.90 |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
you need the model Mistral positions as frontier-class multimodal with an agentic and coding focus, and your evals show it finishing tasks that Large 3 cannot — enough to justify a 5× output rate.
you want Mistral's lowest-cost large model for text generation, summarisation or RAG answers; it is the cheaper of the two on every token and has the same context window.
The output gap ($7.50 vs $1.50) is wider than the input gap, so the more a workload generates, the more Medium 3.5 costs relative to Large 3. A retrieval pipeline that reads 30K tokens and writes 800 is roughly 3× more expensive on Medium 3.5; an agent that writes long code or reports trends toward 5×. Neither model has a cache-hit rate in our table, so repeated system prompts are billed at the full input rate on both.
Because they share the 256K context window and the same API, a common pattern is to default to Large 3 and route only tool-heavy or image-input steps to Medium 3.5. That keeps the expensive output rate confined to the steps that need it. Mistral's pricing page also says batch processing cuts prices by 50%, which applies the same relative saving to both models for offline jobs.
Mistral's pricing page says cached input tokens can reduce input cost by up to 90% for repeated prompts, but it does not list per-model cache rates, so plan on full input pricing until you see the discount on your own invoice.
Mistral's model documentation lists Medium 3.5 and Large 3 as deprecated on Oct 31, 2026, with retirement on Apr 30, 2027 and Mistral Large 4 as the replacement for both. Existing integrations keep working until then, but treat either model as a stop-gap and plan a migration test on Large 4.
No. Medium 3.5 costs $1.50 input and $7.50 output per 1M tokens; Large 3 costs $0.50 and $1.50.
Yes, both list 256K tokens, so long-document capacity is not a reason to pick one over the other.
Mistral's pricing page states that batch processing reduces prices by 50%.
Yes. Mistral deprecated both on Oct 31, 2026 and schedules retirement for Apr 30, 2027, naming Mistral Large 4 as the replacement.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.