Stack
OpenAI
A pay-per-token LLM API serving the GPT model family from a cheap nano tier up to frontier reasoning models.
What it actually does
OpenAI sells access to the GPT model line through an HTTP API and official SDKs. You send text (and images or audio on multimodal models) and pay per token of input and output, with prices set per model rather than per plan. The ladder runs from gpt-5.4-nano at $0.20/$1.25 per 1M tokens up to gpt-5.6-sol at $5/$30, and pro-grade models up to $30/$180.
It is the API most tutorials, courses, and third-party libraries assume by default, so almost any framework or code sample works against it with no adapter. Cost controls include prompt caching (cached input billed at 10% of standard) and a Batch API plus Flex processing that take 50% off input and output for non-urgent jobs.
There is no free tier: you add a card and prepay credits before your first call, so even testing costs money. Priority processing runs 2-2.5x standard rates, and data residency adds 10% on models released after March 2026.
OpenAI vs its main rivals
The tools people actually weigh against OpenAI: Anthropic Claude, Google Gemini, xAI Grok. Same criteria for every column, including where OpenAI loses.
| OpenAI |  Anthropic Claude |  Google Gemini | XxAI Grok | |
|---|---|---|---|---|
Pricing & access | ||||
| Free tier | NoNo | NoOne-time credits | YesOngoing free tier | NoTrial credits |
| Starting price | $0.20 /1M in | $1 /1M in | $0.10 /1M in | $1 /1M in |
Models & capability | ||||
| Tutorials & SDK ecosystem | Largest | Large | Large | Smaller |
| Multimodal (image/audio) | YesYes | YesImage | YesYes | YesImage |
| Frontier reasoning models | YesYes | YesYes | YesYes | YesYes |
Scaling & limits | ||||
| Max context window | 400K | 200K-1M | 1M | 256K |
| Prompt caching discount | Yes10% rate | Yes10% rate | YesPaid only | YesYes |
| Batch 50% discount | YesSelect models | YesYes | YesYes | UnknownUnknown |
Developer & API | ||||
| REST API access | Paid | Paid | Free | Paid |
| Official SDKs | Paid | Paid | Free | Paid |
Data & compliance | ||||
| No training on API data by default | YesYes | YesYes | NoFree tier trains | NoFor credits |
| Data residency option | +10% | US-only +1.1x | Regional | Unknown |
- Widest model ladder for cost routing, from gpt-5.4-nano at $0.20/$1.25 up to frontier tiers in one account
- Prompt caching cuts cached input to 10% of standard, and Batch/Flex take another 50% off for non-urgent jobs
- Largest ecosystem of tutorials, SDKs, and third-party integrations, so most code samples work unchanged
- No API free tier at all, so even testing requires a card and prepaid credits
- Pro-grade models reach $30 input / $180 output per 1M tokens, a trap for solo budgets
- Cheapest nano tier is pricier than Gemini Flash-Lite and Groq for simple high-volume tasks
Pay less for it
5 ways found50% off input and output on non-urgent jobs with a 24h window (select models)
Cached input billed at 10% of the standard rate, big for repeated system prompts
~$2,500 via a Ramp account, more through YC/VC partner tracks, valid ~12 months
Send simple, high-volume tasks to gpt-5.4-nano instead of a flagship
Google Gemini — Flash-Lite at $0.10/$0.40 plus a real free tier covers most simple tasks for a fraction of even OpenAI's nano tier
StackTracker tracks what you actually pay for OpenAI and every other tool, flags overpayment, and shows the dollars you would save by switching.
Is it the right tool for you
- You want the safest default with the most tutorials and library support
- You need one account spanning a cheap nano tier and a frontier reasoning tier
- You will actually implement prompt caching and batching to control spend
- You want to validate an AI feature at $0 before committing → use Google Gemini
- Per-token cost is your top priority on open-model workloads → use Groq or DeepSeek
Track what OpenAI and the rest of your stack cost
StackTracker adds up every subscription, plus your hours, so you see the real number.
Prices and limits last verified 2026-07-20.