These two Mistral budget models share a 256K-token context window but structure their prices differently. Ministral 3 14B charges the same for input and output; Mistral Small 4 is cheaper to read and three times dearer to write.
The cheaper model depends on your input-to-output ratio. Above about 8 input tokens per output token, Mistral Small 4 wins on its lower $0.15 input rate; below that, Ministral 3 14B's flat $0.20 output rate wins. Long-answer generation favours Ministral; document Q&A favours Small 4.
| Ministral 3 14B | Mistral Small 4 | |
|---|---|---|
| Provider | Mistral | Mistral |
| Tier | Budget | Budget |
| Input / 1M tokens | $0.20 | $0.15 |
| Output / 1M tokens | $0.20 | $0.60 |
| Cache hit / 1M | $0.02 | $0.015 |
| Cache write / 1M | — | — |
| Context window | 256K | 256K |
| Notes | Budget multimodal; replaced Pixtral 12B | Mid-tier general; replaced Small 3.1 |
Mistral Small 4 is 25% cheaper on input tokens ($0.15 vs $0.20 per 1M). Ministral 3 14B is 3.0× cheaper on output tokens ($0.20 vs $0.60 per 1M).
| Workload | Ministral 3 14B | Mistral Small 4 | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $4.80 | $5.40 | Ministral 3 14B saves $0.60 |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $6.16 | $4.98 | Mistral Small 4 saves $1.18 |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $0.18 | $0.20 | Ministral 3 14B saves $0.02 |
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.
your workload writes a lot relative to what it reads — drafting, rewriting, chat with long replies — or you need the budget multimodal model that replaced Pixtral 12B.
you feed in long contexts and return short answers, as in retrieval, extraction or classification, where its lower input rate dominates.
Small 4 saves the difference between $0.20 and $0.15 on every 1M input tokens; Ministral saves the difference between $0.60 and $0.20 on every 1M output tokens — eight times as much per token. So Small 4 is cheaper only when you read more than about eight tokens for every one you generate. In the cost table above, the 5:1 support chatbot comes out cheaper on Ministral and the 37:1 document Q&A on Small 4.
Ministral 3 14B replaced Pixtral 12B as Mistral's budget multimodal model, and Small 4 replaced Small 3.1 as the mid-tier general model. If you are on either predecessor, the successor is the one to compare.
Neither model has a cache rate in our table. Mistral's pricing page says batch processing halves prices, which keeps the 8:1 break-even unchanged because both rates fall by the same proportion.
For output, yes ($0.20 vs $0.60 per 1M). For input, Small 4 is cheaper ($0.15 vs $0.20). Which is cheaper overall depends on your ratio of input to output.
Yes, both list 256K tokens.
Ministral 3 14B is listed as Mistral's budget multimodal model, succeeding Pixtral 12B.
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.