OpenAI API pricing

OpenAI runs the widest model lineup of any provider we track, split cleanly into three cost tiers: Budget (the Nano and Mini variants), Standard (the general-purpose Luna and Terra lines, plus the Codex-branded coding variants), and Premium (the Pro/Sol tier and the flagship "GPT Chat Latest"). See the live table above for current rates on every model.

How the tiers actually differ

The gap between OpenAI's Budget and Premium tiers is large — commonly an order of magnitude on both input and output rates — which makes tier selection one of the highest-leverage decisions you'll make, more impactful than most micro-optimizations. A few patterns worth knowing:

• Nano and Mini are built for high-volume, low-complexity work: classification, tagging, short extraction, simple formatting. They're not intended to compete with the Premium tier on hard reasoning, and treating them as a drop-in replacement for complex tasks usually shows up as a quality regression, not a cost saving.
• The Codex-branded variants (GPT-5.2-Codex, GPT-5.3-Codex) are specifically tuned for code generation and agentic coding workflows — if your workload is code-heavy, these are usually a better default than the general-purpose chat variants at a similar price point, even before checking benchmark scores.
• "Pro" and the Sol/flagship tier carry a real premium and are best reserved for tasks where a wrong or shallow answer is expensive — complex multi-step reasoning, high-stakes content, or agentic loops where errors compound across steps.

What to check before committing

OpenAI ships new dated snapshots and renamed variants frequently, so the specific model behind a given name shifts over time — always confirm the current roster on the pricing page rather than assuming a name you remember still points to the same model or price. A few checks worth making every time:

1. Compare Codex variants against general-purpose ones for coding work — the SWE-bench and Aider Polyglot leaderboards will show you whether the coding-tuned variant is actually ahead for your kind of task, since benchmark rank shifts release to release.
2. Don't default to Premium. Route classification and extraction to Nano/Mini first — see How to Cut Your LLM API Bill by 80% for the routing framework — and reserve Premium for the subset of requests that genuinely need it.
3. Check the cache-hit economics. OpenAI's caching mechanics apply the same input/output split logic as every provider — see Prompt Caching Explained for how to structure requests to actually benefit.

When OpenAI vs. a competitor

OpenAI's Standard tier is a reasonable default for general-purpose chat and moderate coding, but it's rarely the outright cheapest option at a given quality bar — Anthropic, Google, and the lower-cost providers (DeepSeek, Qwen, Moonshot) frequently undercut it on pure price for comparable Standard-tier capability. Where OpenAI often earns its price is at the Premium tier for hard reasoning and in the Codex-branded models for coding specifically — check the cost-vs-quality benchmarks before assuming brand recognition equals best value.

Put any two OpenAI models — or an OpenAI model against a competitor — side by side on the compare page, or find the outright cheapest option across every tracked provider on cheapest models.