Every current GPT model with its real per-million-token price — input, output, and cached input — plus a calculator for your own workload and a straight comparison against Claude and Gemini.
| MODEL | INPUT $/1M | CACHED IN | OUTPUT $/1M | BLENDED | CONTEXT | YOUR COST |
|---|---|---|---|---|---|---|
GPT-5.6 Sol gpt-5.6-sol · ≥272K: 2×/1.5× | $4.00 | $0.40 | $20.00 | $8.00 | 1.05M | $14.00 |
GPT-5.6 Terra gpt-5.6-terra · ≥272K: 2×/1.5× | $2.00 | $0.20 | $12.00 | $4.50 | 1.05M | $8.00 |
GPT-5.5 gpt-5.5 · ≥272K: 2×/1.5× | $5.00 | $0.50 | $30.00 | $11.25 | 1.05M | $20.00 |
GPT-5.6 Luna gpt-5.6-luna · ≥272K: 2×/1.5× | $0.20 | $0.02 | $1.20 | $0.45 | 1.05M | $0.80 |
GPT-5.4 Mini gpt-5.4-mini | $0.75 | $0.075 | $4.50 | $1.69 | 400K | $3.00 |
GPT-5.4 Nano gpt-5.4-nano | $0.20 | — | $1.25 | $0.463 | 400K | $0.82 |
USD per 1M tokens from OpenAI’s official pricing page as of August 31, 2026. Blended assumes a 3:1 input:output ratio. Get keys at the OpenAI console.
The GPT-5.6 line is a three-step ladder: Sol is the frontier flagship, Terra the workhorse at half the price, and Luna the fast tier at a twentieth. All three share the same 1.05M-token context window. The older GPT-5.4 Mini and Nano stay on the price list for legacy workloads.
New in the 5.5/5.6 era: a long-context surcharge. Prompts beyond 272K input tokens bill the whole request at 2× input and 1.5× output — and it applies on Batch and Flex too. If you routinely fill a large share of that 1.05M window, price your workload at the surcharge rate.
Cached input is the discount that matters: resend the same prompt prefix and OpenAI bills those tokens at about a tenth of the fresh-input rate (the pricing page now also lists a per-model cache-write fee). For agents and RAG apps with a long, stable system prompt, caching often cuts the input side of the bill by 80–90%.
For anything that can wait, the Batch API processes requests within 24 hours at half price on both input and output, and the Flex service tier gets the same 50% off synchronously when latency doesn’t matter — the cheapest ways to run evaluations, backfills, and bulk classification on GPT.
OpenAI bills per token, with separate input and output rates quoted per million tokens. The current GPT-5.6 line runs from $0.20 input / $1.20 output per 1M tokens on Luna up to $4 / $20 on Sol, with cached input at roughly a tenth of the fresh rate. The table above has every current model.
GPT-5.4 Nano at $0.20 input / $1.25 output per 1M tokens, essentially tied with GPT-5.6 Luna ($0.20 / $1.20). Luna is the better default — same price class, newer model, full 1.05M context, and the deepest cache discount at $0.02 per 1M cached input tokens.
When a request repeats a prompt prefix the API has recently seen (system prompts, few-shot examples, long documents), those tokens bill at the cached rate — about 10× cheaper than fresh input. Caching happens automatically, with a per-model cache-write fee listed on the pricing page. The savings show up in the cached_tokens field of the usage object.
Tier for tier, GPT-5.6 sits between them: Sol ($8/1M blended) costs less than Claude Opus 5 ($10) but more than Gemini 3.1 Pro ($4.50), while Terra ($4.50) lands right between Claude Sonnet 5 ($4) and Gemini 3.7 Flash ($1.50). The matchup table below compares all three, tier by tier.
No — API usage is pay-as-you-go from the first token (new accounts occasionally get trial credits). A ChatGPT Plus or Pro subscription is a separate product and includes no API allowance.
Also see: Claude API pricing · Gemini API pricing · Grok API pricing · DeepSeek API pricing · Kimi API pricing · GLM API pricing · MiniMax API pricing · MiMo API pricing · all 58 models
ReqKey gives every key a credit balance — validate, deduct, enforce limits, and watch per-consumer GPT spend in real time.