OpenAI API Pricing 2026: GPT-4o, o3, o4-mini Costs
Complete per-million-token pricing for every current OpenAI API model — GPT-4o, GPT-4o-mini, o3 and o4-mini. Compare standard, cached and batch rates side by side, then cross-reference against Anthropic, Google and DeepSeek on the main calculator.
Cheapest OpenAI model for coding
GPT-4o-mini — $0.15 input / $0.60output per million tokens. Supports code generation, review and agentic coding at a fraction of GPT-4o's price.
OpenAI pricing per million tokens
USD per 1,000,000 tokens. Cached rate applies to repeated input prefixes; Batch rate requires 24-hour async turnaround.
| Model | Input $/M | Output $/M | Cached $/M | Batch in $/M | Batch out $/M | Context |
|---|---|---|---|---|---|---|
| GPT-4o-mini | $0.15 | $0.60 | $0.075 | $0.075 | $0.30 | 128K |
| o4-mini | $1.10 | $4.40 | $0.55 | $0.55 | $2.20 | 200K |
| GPT-4o | $2.50 | $10.00 | $1.25 | $1.25 | $5.00 | 128K |
| o3 | $10.00 | $40.00 | $5.00 | $5.00 | $20.00 | 200K |
GPT-4o-mini
$0.15 in / $0.60 out · 128K context
Cache saves 50% on input
o4-mini
$1.10 in / $4.40 out · 200K context
Cache saves 50% on input
GPT-4o
$2.50 in / $10.00 out · 128K context
Cache saves 50% on input
o3
$10.00 in / $40.00 out · 200K context
Cache saves 50% on input
OpenAI cost-saving levers
- Prompt caching: repeated prefixes billed at 50% of input — automatic for prefixes of 1,024+ tokens.
- Batch API: exactly 50% off both input and output, in exchange for 24-hour async turnaround.
- Model tiering: route easy traffic to GPT-4o-mini and reserve GPT-4o / o3 for hard prompts — typically cuts total spend 60–80%.
Compare OpenAI against other providers
Frequently asked questions
How much does GPT-4o cost?
GPT-4o lists at $2.50 per million input tokens and $10.00 per million output tokens. With prompt caching enabled, repeated input drops to $1.25/M (50% off); via the Batch API (24-hour turnaround) both input and output halve to $1.25/M input and $5.00/M output.
Is o3 worth it?
OpenAI o3 is the most expensive model in the lineup at $10/M input and $40/M output — roughly 4× GPT-4o. It is worth it only for workloads that genuinely need frontier multi-step reasoning (competition-level math, deep code refactoring, agentic planning). For typical chat and code generation, GPT-4o or o4-mini deliver comparable output at a fraction of the cost.
Does OpenAI offer a batch API discount?
Yes. OpenAI Batch API gives exactly 50% off both input and output pricing in exchange for asynchronous (up to 24-hour) completion. Every model listed here supports it. Batch is ideal for non-latency-sensitive workloads like offline classification, evaluation, and bulk content generation.
Which OpenAI model is cheapest for coding?
GPT-4o-mini is the cheapest OpenAI model that supports code generation, at $0.15 input / $0.60 output per million tokens. For high-volume code workloads where quality is acceptable, it dramatically undercuts GPT-4o.
How does OpenAI prompt caching work?
OpenAI automatically caches repeated input prefixes (minimum 1,024 tokens) and serves cached input at 50% of the standard rate. Caching is applied automatically — no code change required — and is most valuable for long system prompts, few-shot examples, and conversational context that recurs across requests.