Pricing
OpenAI API Pricing in 2026: Every Model, Per 1M Tokens, Dated
OpenAI API pricing per 1M tokens for every current model, from gpt-6-luna at $0.10 input to gpt-6-astra at $10, plus caching rates and a real cost example.
6 min read
Standard model rates from OpenAI’s API pricing page, checked Oct 18, 2026. Batch, Flex, Fast and tool fees re-checked on the same page Oct 19, 2026. Dollars per 1 million tokens unless noted. No affiliate links (how we work).
OpenAI’s pricing page is basically Friends, “The One Where Every Model Has Four Columns”: input, cached input, cache write, output, then a side quest for Batch, Flex and Fast. This post is the OpenAI-only deep dive: every listed Standard rate we verified, the long-context bump, the processing tiers, the tool fees, and one worked monthly bill.
Want the side-by-side? Compare OpenAI, Claude and Gemini API prices. Sibling deep dives: Claude API pricing and Gemini API pricing.
We haven’t tested these models end-to-end. The numbers below are copied from OpenAI’s docs, not from our own usage meters.
TL;DR by situation
- Cheapest listed model: gpt-6-luna at $0.10 input / $0.50 output (Standard, short context).
- Best everyday value: gpt-6.1-sol or gpt-6-sol at $2 / $10. Same headline rates; 6.1-sol’s cached input is cheaper ($0.10 vs $0.20).
- Top of the stack: gpt-6-astra at $10 / $50, jumping to $20 / $75 once a request goes over 272K input tokens.
- Anything that can wait: Batch (or Flex) cuts Standard in half. Fast costs 2× Standard.
- Promo watch: gpt-5.6-sol’s listed rates are promotional “at least through November 21, 2026.”
OpenAI API price table
Standard tier, short context. Headers kept short so the table survives a phone screen.
| Model | Input | Cached | Cache write | Output |
|---|---|---|---|---|
| gpt-6-astra | $10.00 | $1.00 | $12.50 | $50.00 |
| gpt-6.1-sol | $2.00 | $0.10 | $2.50 | $10.00 |
| gpt-6-sol | $2.00 | $0.20 | $2.50 | $10.00 |
| gpt-6-luna | $0.10 | $0.01 | $0.125 | $0.50 |
| gpt-5.6-sol | $4.00 | $0.40 | $5.00 | $20.00 |
| gpt-5.6-terra | $2.00 | $0.20 | $2.50 | $12.00 |
| gpt-5.6-luna | $0.20 | $0.02 | $0.25 | $1.20 |
| gpt-5.5 | $5.00 | $0.50 | — | $30.00 |
| gpt-5.4 | $2.50 | $0.25 | — | $15.00 |
Source: OpenAI “All models” Standard table (pricing, checked Oct 18, 2026). gpt-5.5 and gpt-5.4 rows are the under-272K rates; those two show no cache-write charge. GPT-6 and GPT-5.6 rows charge cache writes at 1.25× input.
This is the GPT-6 / GPT-5.6 / GPT-5.5 / GPT-5.4 slice OpenAI prints on that table. Older families are left out on purpose. OpenAI’s page does not label which models are “current,” so we stick to what’s printed.
Cached input is the discount when the provider reuses a prompt prefix it has already seen. Cache write is what you pay to store that prefix in the first place on the GPT-6 and GPT-5.6 rows.
Long-context and promotional rates
Long context (gpt-6-astra, printed rates): above 272K input tokens, OpenAI lists gpt-6-astra at $20.00 input / $2.00 cached / $25.00 cache write / $75.00 output (checked Oct 18, 2026). That matches the general rule on the same docs: above 272K input, input and cached input go to 2× and output to 1.5× for gpt-6-, gpt-5.6-, gpt-5.5, gpt-5.4 and the pro models. Mini and nano models have no step (notes, multipliers verified Oct 17, 2026).
GPT-5.6 Sol promo: OpenAI’s page states, verbatim, that “GPT-5.6 Sol’s promotional pricing is available at least through November 21, 2026” (reconfirmed Oct 18 and Oct 19, 2026). The Standard row above ($4 / $0.40 / $5 / $20) is that promo price, not a promise of what comes after.
Regional / FedRAMP: endpoints with data residency or FedRAMP carry a 10% uplift for models released on or after March 5, 2026 (same pricing page, Oct 18 / Oct 19).
Batch, Flex and Priority processing
Processing modes multiply the Standard rates. Confirmed again on Oct 19, 2026:
| Mode | vs Standard |
|---|---|
| Batch | 50% of Standard (input, cached, cache write, output) |
| Flex | Same rates as Batch |
| Fast | 2× Standard (renamed from Priority on Jul 30, 2026) |
Examples on gpt-6-astra short context: Batch/Flex $5 / $25; Fast $20 / $100. Use Batch when the job can wait (evals, backfills, overnight summaries). Flex is the same price with different availability trade-offs; we haven’t tested latency differences ourselves. Fast is for when you want Standard quality faster and will pay double.
The page also prints an Ultrafast premium (e.g. gpt-6-astra short context $60 / $300). It is not part of the Batch/Flex/Fast trio above; treat it as a separate speed SKU if your account shows it.
Tool and extra fees
Token fees are not the whole bill. Built-in tools add call or storage charges on top of model rates (pricing, checked Oct 19, 2026):
| Tool | Fee |
|---|---|
| Web search | $10.00 / 1k calls + search content tokens at model rates |
| Image web search | $10.00 / 1k calls + search content tokens at model rates |
| Web search preview (reasoning) | $10.00 / 1k calls + search content tokens at model rates |
| Web search preview (non-reasoning) | $25.00 / 1k calls; search content tokens free |
| File search storage | $0.10 / GB per day (1 GB free) |
| File search tool call | $2.50 / 1k calls (Responses API only) |
| Containers (Shell / Code Interpreter) | 1 GB $0.03 · 4 GB $0.12 · 16 GB $0.48 · 64 GB $1.92 per 20-minute session per container; billed by the minute with a 5-minute minimum |
| Agent Kit ChatKit uploads | $0.10 / GB-day after 1 GB free per account per month |
Tokens for built-in tools still bill at the model rates. OpenAI notes that for some mini models with non-preview web search, search content tokens can bill as a fixed 8,000-input-token block per call. GB means 2^30 bytes.
Worked example: 1M in + 200K out a day
Assume 1,000,000 input tokens and 200,000 output tokens per day, Standard short context, no caching, no tools. Thirty days ≈ one month. Math shown.
gpt-6-luna
- Day: 1 × $0.10 + 0.2 × $0.50 = $0.10 + $0.10 = $0.20
- Month: 30 × $0.20 = $6.00
- Batch month: half → $3.00
gpt-6.1-sol
- Day: 1 × $2.00 + 0.2 × $10.00 = $2.00 + $2.00 = $4.00
- Month: 30 × $4.00 = $120.00
- Batch month: $60.00
gpt-6-astra
- Day: 1 × $10.00 + 0.2 × $50.00 = $10.00 + $10.00 = $20.00
- Month: 30 × $20.00 = $600.00
- Batch month: $300.00
The long-context trap on Astra: a single 300K-token prompt is over 272K, so that prompt’s input bills at $20 per 1M instead of $10. Same tokens, different shape of requests, different bill. Caching flips the story again: pay the cache-write once on GPT-6 / GPT-5.6, then read at the cached rate.
API vs ChatGPT Plus: which one you actually need
These are different products. ChatGPT Plus does not include API credits. ChatGPT plans (Free, Go, Plus, Pro) bill for the chat product; the API bills per token on its own meter. If you mostly talk to ChatGPT in the app, start with the ChatGPT Plus cost breakdown and the ChatGPT plans tracker. If you are wiring models into an app, agent or backend, you are on this page.
Rough rule of thumb from the worked example: light Luna traffic can land under a $20 chat subscription; Astra at a million-plus tokens a day will not. Pick the product that matches how you use the models, not the brand name on the logo.
Using these models through Codex or Copilot instead of the API? See GPT-6.1 Sol vs Terra.
FAQ
Is the OpenAI API free? No free tier is listed on OpenAI’s API pricing page (checked Oct 18–19, 2026).
How much is 1M tokens? It depends on the model: gpt-6-luna is $0.10 in / $0.50 out; gpt-6-astra is $10 / $50 (Standard, short context).
Is cached input cheaper? Yes. On every Standard row above, cached input is a fraction of input (e.g. gpt-6-astra $1.00 vs $10.00).
Does ChatGPT Plus include API credits? No. Plus and other ChatGPT plans do not include API usage; API usage is billed separately.
What’s the cheapest OpenAI model on this table? gpt-6-luna at $0.10 input and $0.50 output (Standard).
Sources
- OpenAI API pricing (Standard model table, long-context astra, GPT-5.6 Sol promo, regional uplift): https://developers.openai.com/api/docs/pricing (checked Oct 18, 2026; tool fees + Batch/Flex/Fast re-checked Oct 19, 2026)
- Batch / Flex / Fast multipliers and long-context step rules: same page; multipliers also noted Oct 17, 2026 in our research log