o3
OpenAI o3 — the highest-capability reasoning model, purpose-built for the hardest problems.
| Type | Price / 1M tokens | Notes |
|---|---|---|
| Input | $10.00 | Per 1M input tokens |
| Output | $40.00 | Per 1M output tokens |
| Cache read | $2.50 | Cached prefix read-back |
At $10/M input and $40/M output, o3 is the most expensive mainstream model. The price is justified only when task accuracy directly drives revenue and Sonnet 5 or o3-mini fail your evals.
Best for
- Frontier mathematics and science research
- Complex multi-step code generation for hard algorithmic problems
- Tasks where wrong answers are very costly (legal, medical, financial analysis)
- Competitive programming and olympiad-level problem solving
Not ideal for
- High-volume workloads (cost is prohibitive)
- Simple tasks any cheaper model handles well
- Latency-sensitive applications (reasoning is slow)
Real cost scenarios
| Scenario | Est. monthly cost | Breakdown |
|---|---|---|
| 1,000 complex research queries/month | $600 | 15k tokens avg per query × 1k = 15M tokens at $10/M input + output |
| Enterprise code generation — 10k complex tasks/month | $8,000 | Heavy reasoning + long output per task |
Estimates with typical token distributions. Use the Prompt Cost Calculator for your exact workload.
How to cut costs on o3
- Always eval o3-mini first — it costs 10× less and handles 70-80% of reasoning tasks at comparable quality.
- Use o3 in a multi-agent setup as the "expert" called only for tasks where cheaper models fail.
- Batch API for o3 gives 50% discount — use for overnight analysis workloads.
Calculate your o3 costs
Other OpenAI models
Compare o3
Frequently Asked Questions
Is OpenAI o3 worth the price?
Only in specific use cases. If your task is genuinely at the frontier of what LLMs can do (research-grade math, cutting-edge code), o3 might be the only model that succeeds. If o3-mini succeeds on your task, use that — it's 9× cheaper on input.
What is the difference between o3 and o3-mini?
o3 is the full flagship reasoning model; o3-mini is a distilled, faster, cheaper version. o3 scores ~5-10% higher on the hardest benchmarks (AIME, competition math). For most production reasoning tasks, o3-mini is sufficient and 9× cheaper.
Pricing verified 2026-08-19 from OpenAI official pricing. All prices in USD per 1M tokens. See methodology and price changelog.