GPT API cost depends on the exact model, input/output mix, cached input, reasoning tokens, tools, and batch eligibility. Use the official live price table as the source of truth and calculate your workload rather than copying a single per-million-token headline.
A practical evaluation method
- Separate input, cached input, output, and non-text tool charges.
- Model p50 and p95 output length, not only averages.
- Track cost per successful task after retries and validation failures.
GPT Pricing at a Glance
OpenAI's current lineup spans a wide price range. Rates per 1M tokens (input / output):
| Model | Input | Output | Best for |
|---|---|---|---|
| GPT-5.5 | $5.00 | $30.00 | Flagship; omnimodal, computer use |
| GPT-5.4 | $2.50 | $15.00 | Balanced reasoning workhorse |
| GPT-5.4-mini | low | low | High-volume, low latency, 400K context |
| GPT-5.4 Nano | $0.20 | — | Cheapest; classification/routing |
| GPT-5.3-codex | — | — | Agentic coding, terminal/computer use |
| GPT-5.4-pro | — | — | Extra-reasoning for high-stakes tasks |
Verify current figures on OpenAI's pricing page before budgeting; the mini/codex/pro tiers vary and change.
The Discounts
- Batch API — 50% off input and output for async jobs.
- Prompt caching cuts repeated input substantially — structure prompts so the stable prefix is reused.
Which GPT Should You Pay For?
- GPT-5.5 for the hardest, most autonomous work — but its $30 output is the priciest in the lineup, so don't default to it.
- GPT-5.4 as the general-purpose default ($2.50/$15) — the price/quality sweet spot.
- GPT-5.4-mini / Nano for classification, routing, and high-volume simple tasks — huge savings.
- GPT-5.3-codex for coding agents specifically (see best coding LLM API).
- GPT-5.4-pro only when extra reasoning compute demonstrably improves accuracy on high-stakes tasks.
Routing simple work down and reserving GPT-5.5 for the hard 5% creates the biggest savings — details in how to reduce AI API costs.
GPT vs the Field
GPT-5.5 ($5/$30) matches Claude Opus 4.8 on input but costs more on output; GPT-5.4 ($2.50/$15) undercuts Claude Sonnet and sits near Gemini 3.1 Pro; all of them are far above DeepSeek V4 Flash ($0.14/$0.28). The full head-to-head is in Claude vs GPT vs Gemini vs DeepSeek; the full rate card in the AI API pricing comparison.
Buying GPT Through a Gateway
If you're routing across providers, LinkModel carries the GPT-5.x family alongside Claude, Gemini, and DeepSeek behind one key, with the price shown before each call. Handy when your app also generates images and video.
Start free with a $1 credit and compare GPT tiers on your own prompts.
