Mistral Small 4 and GPT-6 Luna are both budget models, but Luna is newer and undercuts Small 4 on every listed rate. It also has four times the context window and published cache pricing.
GPT-6 Luna is cheaper on both input and output ($0.10/$0.50 vs $0.15/$0.60 per 1M) and offers 1M context against 256K. Small 4 makes sense mainly if you are standardised on Mistral's platform.
| Mistral Small 4 | GPT-6 Luna | |
|---|---|---|
| Provider | Mistral | OpenAI |
| Tier | Budget | Budget |
| Input / 1M tokens | $0.15 | $0.10 |
| Output / 1M tokens | $0.60 | $0.50 |
| Cache hit / 1M | $0.015 | $0.01 |
| Cache write / 1M | — | $0.125 |
| Context window | 256K | 1M |
| Notes | Mid-tier general; replaced Small 3.1 | New (Sep 29, 2026 DevDay), fastest & cheapest GPT-6-family tier; >272K tokens: $0.20/$0.75, cache $0.02 |
GPT-6 Luna is 33% cheaper on input tokens ($0.10 vs $0.15 per 1M). GPT-6 Luna is 17% cheaper on output tokens ($0.50 vs $0.60 per 1M).
| Workload | Mistral Small 4 | GPT-6 Luna | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $5.40 | $4.00 | GPT-6 Luna saves $1.40 |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $4.98 | $3.40 | GPT-6 Luna saves $1.58 |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $0.20 | $0.15 | GPT-6 Luna saves $0.05 |
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 stack already runs on Mistral — shared billing, existing evals, other Mistral models in the same pipeline — and the per-token difference is small next to the cost of switching.
you are choosing fresh on cost: it is cheaper per token, caches repeated prompts at $0.01 per 1M, and handles prompts up to 1M tokens.
Small 4's 256K limit is enough for most chat and RAG answers, but it rules out single-pass analysis of very long documents, which Luna can take up to 1M. Luna's price does step up above 272K input tokens, to $0.20/$0.75 per 1M, but Small 4 cannot take prompts that long at all.
Our table lists no cache-hit rate for Small 4; Mistral's pricing page says cached input can cut input cost by up to 90% but does not list per-model figures. Luna's cache hit is $0.01 per 1M. Both providers advertise 50% batch discounts, so Luna's lead holds for offline work.
Small 4 replaced Mistral Small 3.1 as Mistral's mid-tier general model, so anyone still on 3.1 should at least move to 4. Luna arrived at OpenAI's Sep 29, 2026 DevDay as the fastest and cheapest GPT-6-family tier.
No. Luna costs $0.10 input and $0.50 output per 1M tokens; Small 4 costs $0.15 and $0.60.
Mistral Small 4 lists 256K tokens and GPT-6 Luna 1M.
Yes. Mistral says batch processing cuts prices by 50%, and OpenAI's Batch API lists Luna at half its standard rates.
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.