Pricing

AI API Pricing in 2026: OpenAI vs Claude vs Gemini (With Calculator)

OpenAI, Claude and Gemini API prices per million tokens, side by side, plus a calculator for your real monthly bill. Every price dated and sourced.

By Engineer Foodie Desk Oct 18, 2026 ✓ Sources checked Oct 18, 2026

8 min read

Prices checked Oct 18, 2026 on OpenAI, Anthropic and Google Gemini. No affiliate links on this page (how we work).

API pricing pages read like the “Severance” board room: lots of numbers, no one explaining what they’re for. So here’s the plain version. Every price is in US dollars per 1 million tokens, taken from each vendor’s official pricing docs and checked on October 18, 2026. You’ll find the links under Sources.

Quick vocabulary. Input is what you send. Output is what the model writes back, and it’s usually the expensive part. Cached input is a repeated prompt prefix, like a system prompt or a stack of docs, that the provider has already seen and bills at a discount.

Disclosure: no affiliate links here, and nobody paid to be listed. Here’s how we work.

TL;DR: cheapest per 1M tokens by use case

Light use (tagging, extraction, quick replies): Gemini 2.5 Flash-Lite is cheapest at $0.10 in and $0.40 out. Claude Haiku 5.5 (prompts up to 100K tokens) and OpenAI’s gpt-6-luna cost $0.10 / $0.50. If your budget is zero, Google’s free tier covers Gemini 3.8 Flash and most of the Flash line.
Mid use (real reasoning, everyday agents): Claude Sonnet 5.5, Claude Sonnet 5 and gpt-6.1-sol all cost $2 in and $10 out.
Heavy use (big models, huge prompts): Claude Opus 5.5 costs $4 / $20, with no long-prompt surcharge up to 1M tokens. GPT-6 Astra costs $10 / $50, rising to $20 / $75 once a request goes over 272K input tokens. Gemini 4 Argon is $2 / $10 until it doubles on Dec 1, 2026.
Anything that can wait: use Batch. All three providers take 50% off.

Calculator

Enter your tokens per request and requests per day, and the calculator estimates your monthly bill for each model, cheapest first. It assumes the standard tier unless you turn on Batch, and it uses the prices dated below.

OpenAI API pricing

Model	Input	Cached input	Output	Notes
gpt-6-astra	$10.00	$1.00	$50.00	Cache write $12.50. Over 272K input: $20 / $75
gpt-6.1-sol	$2.00	$0.10	$10.00	Cache write $2.50
gpt-6-sol	$2.00	$0.20	$10.00	Cache write $2.50
gpt-6-luna	$0.10	$0.01	$0.50	Cache write $0.125
gpt-5.6-sol	$4.00	$0.40	$20.00	Cache write $5.00. Promo price “at least through Nov 21, 2026”
gpt-5.6-terra	$2.00	$0.20	$12.00	Cache write $2.50
gpt-5.6-luna	$0.20	$0.02	$1.20	Cache write $0.25
gpt-5.5	$5.00	$0.50	$30.00	No cache-write charge. Under-272K rate
gpt-5.4	$2.50	$0.25	$15.00	No cache-write charge. Under-272K rate

This is a starting point, not the full lineup. The table covers the GPT-6, GPT-5.6, GPT-5.5 and GPT-5.4 families only. OpenAI’s pricing page doesn’t say which models are current, and GPT-5.2 and older are left out on purpose.

Batch: 50% off input, cached input, cache writes and output, so GPT-6 Astra drops to $5 / $25. Flex costs the same as Batch. “Fast” (called Priority until Jul 30, 2026) costs 2x Standard.
Cache writes: only the GPT-6 and GPT-5.6 models charge for them, at 1.25x input.
Long prompts: above 272K input tokens, input and cached input cost 2x and output costs 1.5x. That applies to gpt-6-, gpt-5.6-, gpt-5.5, gpt-5.4 and the pro models. The mini and nano models don’t have the step.
Free tier: none is listed on the pricing page.

Worked example: GPT-6 Astra, 1M input + 200K output tokens

Input: 1 × $10.00 = $10.00
Output: 0.2 × $50.00 = $10.00
Total: $20.00. With Batch: 1 × $5 + 0.2 × $25 = $5 + $5 = $10.00.
The catch: a single 300K-token prompt goes over 272K, so its input costs 0.3 × $20 = $6.00 instead of 0.3 × $10 = $3.00.

Claude API pricing

Model	Input	Cached input (read)	Output	Cache write (5 min / 1 hr)
Claude Opus 5.5	$4.00	$0.20	$20.00	$5 / $8
Claude Sonnet 5.5	$2.00	$0.10	$10.00	$2.50 / $4
Claude Sonnet 5	$2.00	$0.20	$10.00	$2.50 / $4
Claude Haiku 5.5	$0.10	$0.01	$0.50	$0.125 / $0.20
Claude Fable 5.1	$10.00	$0.25	$50.00	$12.50 / $20

Cache writes and reads: a 5-minute write costs 1.25x input and a 1-hour write costs 2x. A read costs 0.1x input, except Fable 5.1 (0.025x) and Opus 5.5 and Sonnet 5.5 (0.05x).
Batch: 50% off input and output, and it stacks with the caching multipliers. Fast mode can’t be combined with Batch.
Long prompts: Claude 4.6 and later include the full 1M context at the standard rate. The exception is Haiku 5.5: once a prompt goes over 100K tokens, the whole request is billed at $0.50 / $2.50, not just the extra part. Cache reads go to $0.05 and writes to $0.625 / $1.
Free tier: none. New users get “a small amount of free credits to test the API”.
Sonnet 5: the $2 / $10 intro price is now permanent.

Worked example: Haiku 5.5, 1M input + 200K output tokens

Input: 1 × $0.10 = $0.10
Output: 0.2 × $0.50 = $0.10
Total: $0.20, if every prompt stays at or under 100K tokens.
The catch: send the same tokens as five 200K-token prompts and every request crosses the line. That’s 1 × $0.50 + 0.2 × $2.50 = $0.50 + $0.50 = $1.00.
The same job on Sonnet 5: 1 × $2 + 0.2 × $10 = $2 + $2 = $4.00.

Gemini API pricing

Model	Input	Cached input	Output	Notes
Gemini 4 Argon	$2.00	$0.10	$10.00	Doubles Dec 1, 2026, to $4 / $20. Implicit caching only. No free tier
Gemini 3.8 Flash	$0.75	$0.075	$3.75	Batch $0.375 / $1.875. Doubles Jan 1, 2027, to $1.50 / $7.50. Free tier
Gemini 3.6 Flash	$0.75	$0.075	$3.75	Doubles Jan 1, 2027, to $1.50 / $7.50. Free tier
Gemini 3.5 Flash-Lite	$0.30	$0.03	$2.50	Free tier
Gemini 3.1 Flash-Lite	$0.25	$0.025	$1.50	Audio input $0.50. Free tier
Gemini 2.5 Pro	$1.25	$0.125	$10.00	Over 200K: $2.50 / $15 (cached $0.25). Free tier
Gemini 2.5 Flash	$0.30	$0.03	$2.50	Audio input $1.00. Free tier
Gemini 2.5 Flash-Lite	$0.10	$0.01	$0.40	Audio input $0.30. Free tier

Free tier: every model in the table except Gemini 4 Argon is free to use. The trade-off is that free-tier content is used to improve Google’s products.
Long prompts: Gemini 2.5 Pro steps up above 200K tokens. The GA Gemini 3.x and 4 models list no step.
Caching: there’s no write fee. You pay the cached-token price plus storage per 1M tokens per hour: $0.50 for 3.8 and 3.6 Flash (going to $1.00 on Jan 1, 2027), $1.00 for Flash-Lite and 2.5 Flash, $4.50 for 2.5 Pro.
Batch: 50% off on the paid tier, and Flex is also half of Standard.

Worked example: Gemini 3.8 Flash, 1M input + 200K output tokens

Standard: 1 × $0.75 + 0.2 × $3.75 = $0.75 + $0.75 = $1.50
With Batch: 1 × $0.375 + 0.2 × $1.875 = $0.375 + $0.375 = $0.75
Free tier: $0, as long as you’re fine with Google using the content.
The catch: prices double on Jan 1, 2027. New total: 1 × $1.50 + 0.2 × $7.50 = $1.50 + $1.50 = $3.00.

What drives your bill

Output, not input. On every model in these tables, output costs four to more than eight times as much as input. Long answers cost more than long questions.
Caching. If you resend the same long prefix, pay once to write it and then read it at a fraction of the input price.
Batch. Half price for jobs that can wait: evals, backfills, nightly summaries.
Long-context steps. OpenAI charges more above 272K input tokens, Haiku 5.5 above 100K, and Gemini 2.5 Pro above 200K.
Promo end dates. gpt-5.6-sol is promo-priced “at least through Nov 21, 2026”. Gemini 4 Argon doubles Dec 1, 2026, and 3.8 and 3.6 Flash on Jan 1, 2027.

API vs a $20 subscription

These are separate products. ChatGPT Plus and Pro don’t include API access, so your API usage is billed on its own. Claude Max is the exception: Max 5x and Ultra (formerly Max 20x; claude.com/pricing, verified Nov 1, 2026) include $100 and $200 a month in Claude API credits, which don’t cover Claude Code. If you mostly chat and only occasionally script something, a subscription is usually the calmer choice, and our ChatGPT vs Claude comparison helps you pick one. Consumer plan prices are in our ChatGPT, Claude and Gemini trackers, and how much ChatGPT Plus costs breaks that one down.

Here’s what I’d do

Run it like a kitchen in “The Bear”: you don’t put the expensive cut in every dish.

Prototype on Gemini’s free tier with non-sensitive data, and measure your real token counts before you commit to anyone.
Send the cheap work to a small model like Gemini 2.5 Flash-Lite, gpt-6-luna or Haiku 5.5, and keep Haiku prompts at or under 100K tokens.
Escalate on purpose to Sonnet 5.5, Opus 5.5 or GPT-6 Astra, and only for the hard reasoning.
Batch anything that can wait. It’s half price everywhere.
Cache long prefixes, and remember that GPT-6 and GPT-5.6 charge for the write.
Calendar the dates: Nov 21, Dec 1, Jan 1.

For coding specifically, see Claude Code pricing, Cursor vs Claude Code and our AI coding tools comparison.

FAQ

Is the OpenAI API free? No, OpenAI’s pricing page lists no free tier for the API.

Is the Claude API cheaper than OpenAI? It depends on the model. Haiku 5.5 and gpt-6-luna tie at $0.10 / $0.50, while Opus 5.5 ($4 / $20) undercuts GPT-6 Astra ($10 / $50).

What is a token? A token is a small chunk of text, often part of a word, and APIs bill per million of them.

Does ChatGPT Plus include API access? No, ChatGPT Plus and Pro don’t include API usage, which is billed separately.

Is there a Gemini API free tier? Yes, most Gemini Flash, Flash-Lite and 2.5 models are free, but Gemini 4 Argon isn’t.

Sources

OpenAI API pricing: https://platform.openai.com/docs/pricing (checked Oct 18, 2026)
Anthropic Claude pricing: https://docs.anthropic.com/en/docs/about-claude/pricing (checked Oct 18, 2026)
Claude Max API credits: https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans (checked Oct 18, 2026)
Google Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing (checked Oct 18, 2026)