OpenAI API Pricing 2026: What GPT-6 Astra, Sol, Terra and Luna Cost per Token and per Task
AIHub
OpenAI API pricing in 2026 runs from 0.10 dollars to 10 dollars per million input tokens and from 0.50 to 50 dollars per million output tokens on the current lineup. GPT-6 Luna costs $0.10 in and $0.50 out, GPT-6.1 Terra $1 and $5, GPT-6.1 Sol $2 and $10, and GPT-6 Astra $10 and $50. Output costs five times input on all four. Cached input drops to between 5 and 10 percent of the normal rate, Batch and Flex halve everything, and any request over 272,000 input tokens is billed at a higher long-context rate from the first token.
Those are list prices from OpenAI’s own pricing page and model pages, read on 2026-11-16. Per-token prices are hard to budget with, so this guide also prices six real jobs on each model, from a single question to a 30-step coding agent task, explains the two charges that surprise people most (reasoning tokens and the long-context step), and shows where the API beats a 20 dollar ChatGPT Plus subscription.
OpenAI API prices per million tokens
| Model | Input | Output | Cached input | Cache write (30 min) | Cache write (12 h) |
|---|---|---|---|---|---|
| GPT-6 Astra | $10 | $50 | $1.00 | $12.50 | $25 |
| GPT-6.1 Sol | $2 | $10 | $0.10 | $2.50 | $5 |
| GPT-6.1 Terra | $1 | $5 | $0.05 | $1.25 | $2.50 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 | $0.125 | $0.25 |
US dollars per million tokens, Standard tier, prompts up to 272,000 input tokens, from OpenAI’s API pricing page on 2026-11-16. A 30-minute cache write costs 1.25 times the input price and a 12-hour write 2.5 times. Cached reads are 5 percent of input on Sol, Terra and Luna and 10 percent on Astra.
Astra is the top model, Sol is OpenAI’s own pick for complex coding, computer use and professional work (“near-Astra performance at a lower cost”, in the words of its model page), Terra is the new middle tier released at the end of October, and Luna is the cheap, fast model for classification, extraction and short copy. All four have a 1,050,000-token context window.
What common jobs cost on each model
The token counts below are the ones used across this site’s guides, so the numbers line up with Claude API pricing, ChatGPT alternative for writing and ChatGPT alternative for coding. Standard tier, no reasoning tokens counted, no caching unless stated.
| Job (tokens in / out) | GPT-6 Luna | GPT-6.1 Terra | GPT-6.1 Sol | GPT-6 Astra |
|---|---|---|---|---|
| Short question (1,000 / 500) | $0.0004 | $0.0035 | $0.007 | $0.035 |
| Ten-turn conversation (39,000 / 5,000) | $0.006 | $0.064 | $0.13 | $0.64 |
| 1,500 word article with one edit (4,000 / 4,000) | $0.002 | $0.024 | $0.048 | $0.24 |
| Summary of a 100-page PDF (65,000 / 1,000) | $0.007 | $0.07 | $0.14 | $0.70 |
| 30-step coding agent task (600,000 / 30,000) | $0.08 | $0.75 | $1.50 | $7.50 |
| Same agent task, 80% of input cached | $0.03 | $0.29 | $0.59 | $3.18 |
The ten-turn conversation costs far more than ten short questions because every turn resends the earlier ones: 300-token prompts and 500-token answers add up to 39,000 input tokens by the end. The agent task is the same effect at scale. An agent rereads the files and its own history on every step, so input dominates the bill, and that is exactly the input caching discounts. The cached row counts cache reads only; the first write of each prefix adds a little at 1.25 times the input rate.
The steps between models are clean. Terra is exactly half of Sol on every job, Sol is a fifth of Astra, and Luna is a tenth of Terra. That makes routing easy to reason about: a job that Terra does as well as Sol is a 50 percent saving with no other change.
Batch, Flex, Fast and Ultrafast
OpenAI sells the same models at four speeds. The price moves by a fixed multiple of the Standard rate, for input, output and cache alike.
| Tier | Multiple of Standard | GPT-6.1 Sol in / out | GPT-6 Astra in / out | What you trade |
|---|---|---|---|---|
| Batch | 0.5x | $1 / $5 | $5 / $25 | results within 24 hours, asynchronous |
| Flex | 0.5x | $1 / $5 | $5 / $25 | slower responses and occasional resource-unavailable errors |
| Standard | 1x | $2 / $10 | $10 / $50 | the default |
| Fast | 2x | $4 / $20 | $20 / $100 | faster output; this was called Priority processing until 2026-07-30 |
| Ultrafast | 6x | $12 / $60 | $60 / $300 | the fastest output, Sol and Astra only |
Batch is the one most teams should use more. Summarising 1,000 PDFs on GPT-6.1 Sol costs about 140 dollars at Standard rates and about 70 through the Batch API, and batch stacks with cached input. Flex gives the same half price for synchronous calls when a slow or failed response is acceptable, which suits evaluation runs and background enrichment. Fast and Ultrafast are for user-facing latency; at six times the price, Ultrafast on a 30-step Sol agent task is 9 dollars instead of 1.50.
Two surcharges sit on top of every tier. Regional processing (data residency) endpoints cost 10 percent more for models released on or after 2026-03-05, which covers the whole current lineup, and so do FedRAMP endpoints.
The 272K long-context step
All four current models accept prompts of over a million tokens, but the price changes at 272,000 input tokens. Above that line the whole request, not just the excess, is billed at twice the input and cache rates and 1.5 times the output rate.
| Model | Input over 272K | Output over 272K | Cached input over 272K |
|---|---|---|---|
| GPT-6 Astra | $20 | $75 | $2.00 |
| GPT-6.1 Sol | $4 | $15 | $0.20 |
| GPT-6.1 Terra | $2 | $7.50 | $0.10 |
| GPT-6 Luna | $0.20 | $0.75 | $0.02 |
The step creates a cliff worth engineering around. On GPT-6.1 Sol a 250,000-token prompt with a 2,000-token answer costs about 52 cents. Add 50,000 more tokens and the same request costs about 1.23 dollars, more than double for 20 percent more input. If you send whole repositories or document sets, retrieve the relevant parts instead, or split the job, and keep each request under 272K. Agent loops rarely hit the line, because each step is a separate request; single-shot document analysis hits it often.
Reasoning tokens: the cost you do not see
Every current OpenAI model reasons before it answers, and the reasoning is billed as output tokens even though the API does not return the text. GPT-6.1 Sol takes a reasoning.effort setting of low, medium (the default), high, xhigh or max, and higher effort means more hidden output. A short question that returns 500 visible tokens after 2,000 reasoning tokens costs about 2.7 cents on Sol, nearly four times the 0.7 cents in the job table.
That is why the tables above are floors. Before budgeting, run a sample of real requests at the effort level you plan to use and read the reasoning token count in each response’s usage object. For extraction, classification and formatting jobs, low effort is usually enough and keeps the bill close to the table. OpenAI’s token counting guide covers the visible side; only real runs show the reasoning side.
Older OpenAI models still on sale
Old model IDs keep working, and several cost more than their replacements.
| Model | Input | Output | Note |
|---|---|---|---|
| GPT-6 Sol | $2 | $10 | same price as 6.1 Sol, but cached input is $0.20 instead of $0.10 |
| GPT-5.6 Sol | $4 | $20 | promotional price until 2026-11-30 |
| GPT-5.6 Terra | $2 | $12 | 2x the input and 2.4x the output price of 6.1 Terra |
| GPT-5.6 Luna | $0.20 | $1.20 | 2x the input and 2.4x the output price of 6 Luna |
| GPT-5.5 | $5 | $30 | |
| GPT-5.4 | $2.50 | $15 | |
| GPT-5.4 mini | $0.75 | $4.50 | |
| GPT-5.4 nano | $0.20 | $1.25 | |
| GPT-5 | $1.25 | $10 | |
| GPT-5 mini | $0.25 | $2 | |
| GPT-5 nano | $0.05 | $0.40 | the cheapest OpenAI text model still listed |
| GPT-4.1 | $2 | $8 | |
| GPT-4o | $2.50 | $10 | |
| GPT-4o mini | $0.15 | $0.60 | |
| GPT-5.5 Pro | $30 | $180 | with GPT-5.4 Pro, the most expensive OpenAI text model listed |
Standard rates per million tokens for OpenAI’s own endpoint, as listed in OpenRouter’s catalog on 2026-11-16; the GPT-5.6 Sol row was confirmed on OpenAI’s model page. An application pinned to GPT-5.5 pays 5 and 30 dollars for work GPT-6.1 Sol does at 2 and 10, so checking the model IDs in production is often the cheapest price cut available. Fine-tuning is closed to new users, and the remaining fine-tunable models (GPT-4.1 and GPT-4.1 mini) are deprecated.
Tools, images and voice
Built-in tools add a per-call fee on top of tokens. Web search is 10 dollars per 1,000 calls, plus the retrieved content billed as input at the model’s rate. File search storage is 0.10 dollars per GB per day after the first free GB, and file search calls in the Responses API are 2.50 dollars per 1,000. Hosted Shell and Code Interpreter containers cost 3 cents per 20-minute session at 1 GB of memory, rising to 1.92 dollars at 64 GB.
Image generation with gpt-image-2.6 bills image output at 30 dollars per million image tokens (15 through Batch), and OpenAI’s image calculator converts that into a per-image price by size and quality. Voice runs through gpt-realtime-2.1 at 32 dollars per million audio input tokens and 64 per million audio output, or gpt-realtime-2.1-mini at 10 and 20; GPT-Live voice sessions are 5 cents a minute, billed per second, plus the backend model’s tokens.
OpenAI against Anthropic, Google and the market
| Model | Input | Output |
|---|---|---|
| GPT-6.1 Sol | $2 | $10 |
| Claude Sonnet 5.5 | $2 | $10 |
| Gemini 3.1 Pro | $2 | $12 |
| Grok 4.7 | $2 | $6 |
| GPT-6.1 Terra | $1 | $5 |
| GPT-6 Luna | $0.10 | $0.50 |
| Claude Haiku 5.5 | $0.10 | $0.50 |
| Gemini 3.5 Flash Lite | $0.30 | $2.50 |
OpenAI prices read 2026-11-16; the others are vendor list prices from the AI model API pricing data checked 2026-11-02. At the frontier OpenAI, Anthropic and Google all charge 2 dollars in, and Luna and Haiku 5.5 match to the cent. Terra has no direct equivalent from Anthropic or Google at 1 and 5 dollars, which makes it the obvious first model to test when Sol is good enough but too expensive. The top tiers also line up: GPT-6 Astra and Claude Fable 5.1 are both 10 and 50.
Against the whole market OpenAI is mid-to-expensive. Across 335 paid models on OpenRouter, checked 2026-11-02, the median price is $0.40 in and $1.88 out. OpenAI has the largest catalog of any provider in that set (58 models) at a median of $2 in and $10 out, and 81 percent of its models offer cached input. The full distribution, by provider and by context window, is in the LLM API pricing statistics.
OpenAI API or ChatGPT Plus?
ChatGPT Plus is 20 dollars a month. Using the job table, 20 dollars at API rates buys about 2,850 short questions on GPT-6.1 Sol, about 156 ten-turn conversations (313 on Terra), or about 13 uncached 30-step agent tasks (34 with caching). If you have fewer than about five real conversations a day, paying per token is cheaper. If you live in ChatGPT, use image generation and voice, or want deep research, Plus wins, because none of those come with an API key.
Two details change the sum for heavier users. ChatGPT Pro (100, 200 and 500 dollar tiers) includes the same amount in monthly API credits, and Business plans include 20 dollars of API credit per Standard seat and 100 per Premium seat, pooled and capped at 500 a month per workspace. Plus includes none. Unused credits expire at the end of each billing cycle.
The plan side, including Plus’s context limits and how it compares with Claude Pro at the same price, is in Claude Pro vs ChatGPT Plus. If you would rather pay per message without managing API keys, AI chat aggregators explain the four ways those apps bill. AIHub is one of them, so read that with the disclosure in mind: it charges GPT-6 Sol, GPT-6 Luna and GPT-5.4 mini at OpenAI’s list price in credits, alongside Claude and Gemini, on one bill.
How to keep an OpenAI API bill small
- Start new work on GPT-6.1 Terra and move a task to Sol only when Terra’s answer is measurably worse. The price gap is exactly 2x.
- Send classification, extraction and routing to GPT-6 Luna. A thousand short calls cost about 35 cents.
- Set reasoning effort per task. Low effort on simple jobs keeps hidden output tokens from multiplying the bill.
- Keep single requests under 272,000 input tokens. Crossing the line doubles the input rate for the whole request.
- Put stable content (system prompt, tool definitions, reference documents) at the start of the prompt so it is served from the cache at 5 percent of the input price.
- Run anything that can wait through Batch, and anything that can tolerate a slow reply through Flex. Both are half price.
- Check the model IDs in production. GPT-5.5 and GPT-5.6 Terra cost more than the models that replaced them.
Prices change, and OpenAI runs promotions with end dates. Every figure here is dated 2026-11-16; check OpenAI’s pricing page before signing a contract or launching a feature.