OpenAI API Pricing 2026: Every Model, Tier and Fee

Last Updated: November 1, 2026

OpenAI API pricing runs from $0.10 per million input tokens and $0.50 per million output tokens for GPT-6 Luna up to $10 and $50 for GPT-6 Astra, its most capable model. In between sit GPT-6.1 Terra at $1 and $5 and GPT-6.1 Sol at $2 and $10. You pay only for what you use, from prepaid credits, and the Batch API or Flex processing cuts those rates in half.

This guide covers every OpenAI API model family, the discounts and surcharges, tool fees, fine-tuning, usage tiers and billing, a price comparison with Claude, Gemini, Grok and DeepSeek, and worked monthly cost examples.

OpenAI's four flagship GPT-6 models cost between $0.10 and $10 per million input tokens and between $0.50 and $50 per million output tokens. Output always costs five times as much as input on these models, so the length of the answers you generate drives most of your bill.

Prices come from OpenAI's API pricing page, checked on November 1, 2026. All prices are in US dollars per million tokens (1M tokens is roughly 750,000 English words, since OpenAI's token guide puts 100 tokens at about 75 words).

All four flagship models share a 1,050,000-token context window and can return up to 128,000 output tokens, according to OpenAI's model comparison page. OpenAI released GPT-6 Astra on September 3, GPT-6 Sol and Luna on September 22, GPT-6.1 Sol on September 29 and GPT-6.1 Terra on October 26, 2026, per the API changelog.

Two surcharges catch people out. On Astra, Sol and Terra, a prompt longer than 272,000 input tokens is billed at 2x the input rate and 1.5x the output rate for the whole request, according to each model's page. And data residency (regional processing) adds 10% for models released on or after March 5, 2026.

No separate line item exists, but you still pay for them. Every GPT-6 model reasons before it answers, and OpenAI's reasoning guide says those hidden reasoning tokens "are billed as output tokens." You cannot read them, but the usage data in each response shows how many you paid for.

The reasoning effort setting is your main dial. GPT-6.1 Sol supports low, medium (the default), high, xhigh and max, and OpenAI's GPT-6 guide suggests starting at medium and raising it only if the quality gain is worth the extra tokens. The older o-series models no longer appear on the main pricing page.

Cached input is up to 95% cheaper than normal input, Batch and Flex cost 50% less than standard rates, and Fast mode costs twice as much.

Caching is automatic. When a prompt starts with the same 1,024 or more tokens as a recent request, OpenAI bills that repeated part at the cached rate. OpenAI's prompt caching guide says cache reads cost 0.1x the normal input price on GPT-5.6 and later models, and 0.05x on GPT-6.1 Sol and Terra.

Writing to the cache costs 1.25x normal input for the default 30-minute cache and 2.5x for the 12-hour cache (Responses API only). Each input token is billed once, at the uncached, cached or cache-write rate.

The Batch API gives a 50% discount in exchange for results within 24 hours instead of seconds. One batch can hold up to 50,000 requests in a file of up to 200 MB.

Flex processing charges Batch rates but keeps the normal request-and-response flow. You set "service_tier" to "flex", accept slower answers, and occasionally get a "429 Resource Unavailable" error, which OpenAI does not charge for. It is still in beta with limited model support.

Fast mode replaced Priority processing on July 30, 2026. It costs 2x standard rates and runs up to 2.5x faster, per OpenAI's Fast mode guide. Requests still tagged "priority" are routed to Fast mode automatically. The pricing page also lists an Ultrafast tier for GPT-6 Astra and GPT-6.1 Sol at $60 input and $300 output for Astra, and $12 and $60 for Sol.

Figures are input / output. Cached input discounts apply on top of every tier.

Image models cost $30 per million output tokens, realtime voice runs $32 to $64 per million audio tokens on the full model, transcription starts at $0.0045 a minute, and embeddings cost $0.02 to $0.13 per million tokens. Moderation is free.

Specialized models include chat-latest at $5 input and $30 output, gpt-5.3-codex at $1.75 and $14, and GPT-5.6 Sol at a promotional $4 and $20 through at least November 21, 2026.

Web search costs $10 per 1,000 calls plus content tokens at your model's rate, file search costs $2.50 per 1,000 calls plus storage, and Code Interpreter runs in containers priced by size.

Fees add up in agents: 5 web searches per task is $50 per 1,000 tasks in call fees before any tokens. The Assistants API shut down on August 26, 2026, so new builds should use the Responses API.

OpenAI is winding down fine-tuning. Its pricing page says the platform is no longer open to new users, and existing users can create training jobs only for a few more months. Fine-tuned models keep working until their base models are deprecated.

The one model still priced is o4-mini: training costs $100 per hour, and inference on the fine-tuned model costs $4 input and $16 output per million tokens. Sharing your training data with OpenAI halves inference to $2 and $8. The GPT-6 model pages we checked list fine-tuning as not supported.

OpenAI has one free tier and three paid tiers, Build, Launch and Grow, and you move up automatically as your total credit purchases grow. OpenAI cut the tiers from five to three on October 6, 2026.

The figures come from OpenAI's rate limits guide and model pages. RPM is requests per minute and TPM is tokens per minute. Since the free tier does not support the GPT-6 flagships, you need at least a $5 purchase to build with them.

The OpenAI API is prepaid. You buy credits (minimum $5), usage draws down the balance, and requests stop when it reaches $0. Purchased credits expire after one year and cannot be refunded.

OpenAI's prepaid billing article also describes auto-recharge, which tops up your balance by a set amount (minimum $5) when it falls below a threshold you choose. You can set spend limits per organization or project on top of your tier's monthly cap.

API billing is separate from ChatGPT. OpenAI's billing help page states that "ChatGPT and the API platform have separate billing systems," so a Plus or Pro subscription does not pay for API calls. If you want the consumer plans instead, see our ChatGPT pricing guide.

A customer-support chatbot handling 10,000 conversations a month costs about $7 on GPT-6 Luna, $72 on GPT-6.1 Terra, $143 on GPT-6.1 Sol and $730 on GPT-6 Astra, using the assumptions below. Summarizing 1,000 long documents costs between $0.75 and $150 depending on the model and whether you use Batch.

Assume each conversation uses 5,000 input tokens and 1,000 output tokens (including reasoning). Of the input, 3,000 tokens are the same instructions and help-center text every time, so they bill at the cached rate. The other 2,000 are the customer's messages. Over a month that is 30M cached tokens, 20M fresh input tokens and 10M output tokens.

Caching does real work here. Without it, GPT-6.1 Sol would bill all 50M input tokens at $2, or $100 for input and $200 in total, so caching saves $57 a month (about 29%). Cache-write charges add a small amount when the cache refreshes, and this example ignores them.

Assume 1,000 reports of about 7,500 words each (10,000 tokens), with 1,000 output tokens per summary. That is 10M input tokens and 1M output tokens. Because nobody needs the summaries in seconds, this is a textbook Batch job.

Terra at $7.50 versus Astra at $75 is a tenfold gap, so test the cheaper model on 20 or 30 documents before moving up.

At the top and mid tiers, OpenAI matches Anthropic almost exactly and Google's Gemini 4 Argon for now, while xAI's Grok 4.7 has cheaper output and DeepSeek is the cheapest per token. At the budget end, GPT-6 Luna and Claude Haiku 5.5 are tied at $0.10 and $0.50.

Sources: Anthropic pricing, Gemini API pricing, xAI API pricing and DeepSeek pricing, all checked November 1, 2026.

A few details change the math:

Applied to the 10,000-conversation chatbot above, the mid-tier bills come out at $143 on GPT-6.1 Sol, $143 on Claude Sonnet 5.5 (its $0.10 cache-hit rate matches Sol), $115 on Grok 4.7 and about $67 on DeepSeek V4 Pro at peak rates. For the bigger trend behind these numbers, read our report on AI token prices hitting record lows.

Start with GPT-6.1 Terra or GPT-6 Luna for most business apps, move to GPT-6.1 Sol for complex coding and agents, and reserve GPT-6 Astra for tasks where a wrong answer is expensive. Test on your own data before committing, because the price gap between tiers is up to 100x.

Whatever you choose, put fixed instructions at the start of every prompt so they cache, send anything that can wait to Batch or Flex, and keep reasoning effort as low as quality allows.

Not for the current flagship models. OpenAI has a free tier with a $100 monthly usage limit, but GPT-6 Astra, Sol, Terra and Luna are not supported on it. You need to buy at least $5 in credits to reach the Build tier, though the moderation endpoint is free for all API users.

GPT-6 Luna is the cheapest flagship model at $0.10 per million input tokens and $0.50 per million output tokens. Batch or Flex drops it to $0.05 and $0.25.

No. OpenAI says ChatGPT and the API platform have separate billing systems, so API usage is paid for with prepaid API credits, not your ChatGPT subscription.

The Batch API cuts input and output prices by 50% in exchange for results within 24 hours.

Yes. Purchased API credits expire one year after purchase and are non-refundable, according to OpenAI's prepaid billing help article. When your balance hits $0, API requests stop unless auto-recharge is on.

OpenAI API pricing in November 2026 comes down to four models: Luna at $0.10 and $0.50, Terra at $1 and $5, Sol at $2 and $10, and Astra at $10 and $50 per million tokens. Caching, Batch and Flex cut those numbers; Fast mode, long prompts and data residency raise them. OpenAI is priced in line with Anthropic, below Gemini once Google's December increase lands, and above Grok and DeepSeek on raw per-token cost. Run your own workload through two models before you commit, and recheck OpenAI's pricing page every few weeks, because these numbers change often.

By AI Business Weekly Editorial

Written by the AI Business Weekly team with AI assistance, then reviewed, fact-checked against cited sources and edited by Sameer Khan.

About the editor: Sameer Khan is the founder of AI Business Weekly. He has nearly a decade of B2B sales experience across SaaS, research and advisory, and AI, working with HR leaders and executives on research and AI adoption. He holds an MBA from the Ted Rogers School of Management.