Table of Contents
Key Takeaway
- 💰 AI API pricing is a two-horse race at the frontier: OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 both price at $10 per million input and $50 output, while Google’s Gemini 4 Carbon undercuts at $2/$10 until November 30, 2026.
- 📅 Google’s floor moves twice: Gemini 4 Carbon and Argon double to $4/$20 on December 1, 2026, and Gemini 3.8 Flash doubles to $1.50/$7.50 on January 1, 2027.
- 🪙 The small tier is where budgets are won: Anthropic’s Claude Haiku 5.5 runs a 10-million-token month for $0.75 against OpenAI GPT-6 Luna’s $1.50 and Google’s Flash-Lite at $4.00.
- 🧠 Output is always the expensive half: every provider charges 5 to 10 times more per million output tokens than per million input tokens, and the batch lane halves the whole bill at all three.
AI API pricing shifted twice in November 2026, and both moves favor whoever reads the tables. OpenAI published a full GPT-6 price structure that runs from $0.10 per million input tokens on the smallest model to $10 on the frontier, Anthropic lists identical frontier numbers a few pages away, and Google quietly posted a price increase scheduled for December 1 that doubles its strongest reasoning model. If your product bills monthly, the difference between the three providers is no longer a rounding error: on a realistic 10-million-token month for a small-model workload, the same bill varies from $0.75 to $4.00 depending on which provider you picked. This page compares the actual posted rates across the Big Three, tier by tier, with worked monthly figures computed directly from each provider’s own price list, so the quotes you write survive the next price change.
AI API pricing only looks like three simple columns; the calendars underneath them are what move your bill. The rates below are the posted 2026 list prices, stated the way the providers state them: per million tokens, input and output separately. Each figure links back to the pricing page we pulled it from, and every worked cost in this article is arithmetic on those posted numbers, not a vendor guess.
AI API Pricing on the Frontier Tier: November 2026 Rates
The frontier tier is what you pay when you want the best reasoning a provider sells:
| Provider / model | Input per 1M | Output per 1M | Effective Dec 1, 2026 |
|---|---|---|---|
| OpenAI gpt-6-astra | $10.00 | $50.00 | unchanged |
| Anthropic Claude Fable 5.1 | $10.00 | $50.00 | unchanged |
| Google Gemini 4 Carbon | $2.00 | $10.00 | $4.00 / $20.00 |
Two of the three are perfectly synchronized. OpenAI’s Astra and Anthropic’s Fable 5.1 both bill $10 per million input and $50 per million output, which makes provider switching a capability question rather than a price question at the top tier.
Google’s Carbon enters at a quarter of that number, and Google’s own page marks the increase: $2.00 input and $10.00 output hold through November 30, 2026, and the price doubles to $4.00 and $20.00 starting December 1, 2026. The same page applies the identical move to Gemini 4 Argon. For readers tracking context costs, Astra’s long-context window bills at $20 input and $75 output, and it is the only model in the frontier race whose long-context figures diverge that far from its short-context ones.
The Middle Tier: Where Most Production Builds Live
The middle tier is where AI API pricing stops being news and starts being a maintenance line, because this is where production volume actually runs:
| Provider / model | Input per 1M | Output per 1M | Cached input |
|---|---|---|---|
| OpenAI gpt-6.1-sol | $2.00 | $10.00 | $0.10 |
| Anthropic Claude Sonnet 5.5 | $2.00 | $10.00 | $0.10 (hits) |
| Google Gemini 4 Hydrogen | $1.50 | $7.50 | $0.075 |
| Google Gemini 3.8 Flash | $0.75 | $3.75 | $0.075 |
OpenAI’s gpt-6.1-sol and Anthropic’s Sonnet 5.5 post identical $2 and $10 rates, and Google splits the tier into two answers: Hydrogen for heavy everyday work at $1.50 and $7.50, and Gemini 3.8 Flash at $0.75 and $3.75 for long-horizon engineering jobs. Google’s Flash rates carry their own calendar: $0.75 and $3.75 hold through December 31, 2026, and both double on January 1, 2027. The cached-input column is the quiet weapon in this tier. When your requests repeatedly ship the same system prompt and knowledge base, OpenAI prices that repeated prefix at $0.10 per million against $2.00 fresh, a 20-fold cut, and Sonnet 5.5’s cache hits match it exactly. We mapped that convergence across providers in the $0.10 cache convergence, and it remains the fastest cost lever on the page for teams that bill month over month.
The Small-Model Tier: Where the Margins Are Decided
Small models carry classification, extraction, routing, and summarization, and their rates define what a high-volume product can afford to charge its own customers. The cheapest AI API pricing on the Big Three lives in this tier:
| Provider / model | Input per 1M | Output per 1M |
|---|---|---|
| OpenAI gpt-6-luna | $0.10 | $0.50 |
| Anthropic Claude Haiku 5.5 | $0.05 (up to 100k prompts) | $0.25 |
| Google Gemini 3.1 Flash-Lite | $0.25 | $1.50 |
| Google Gemini 3.5 Flash-Lite | $0.30 | $2.50 |
Haiku 5.5 is the price point to understand here.
Anthropic bills its high-volume model at $0.05 per million input tokens for prompts up to 100,000 tokens and $0.25 for output, and steps up to $0.25 input and $1.25 output only for prompts over 100,000.
OpenAI’s Luna counters with a 5:1 output ratio at $0.10 and $0.50, and Google’s Flash-Lite models run two to three times pricier on both axes. One more structural note: Anthropic’s tier splits by prompt size, while OpenAI and Google split by context length, so the comparison is only apples to apples inside your own prompt-size envelope.
What 10 Million Tokens a Month Actually Costs
Multiply the posted rates by a realistic workload of 10 million input and 1 million output tokens per month, and the tier table becomes a bill:
| Model (standard tier) | Monthly cost |
|---|---|
| Anthropic Claude Haiku 5.5 | $0.75 |
| OpenAI gpt-6-luna | $1.50 |
| Google Gemini 3.1 Flash-Lite | $4.00 |
| Google Gemini 4 Hydrogen | $22.50 |
| OpenAI gpt-6.1-sol / Anthropic Claude Sonnet 5.5 | $30.00 |
| Google Gemini 4 Carbon | $30.00 now, $60.00 from Dec 1 |
| OpenAI gpt-6-astra / Anthropic Claude Fable 5.1 | $150.00 |
This table is AI API pricing turned into an actual bill. Every row is straight arithmetic on posted rates: ten million input tokens plus one million output tokens at each model’s input and output price. The spread is the story. The same workload can cost as little as $0.75 a month or as much as $150, and the frontier decision doubles again the day Google’s December increase lands. At the mid tier a heavier agent day, five million input and half a million output, bills $15.00 on gpt-6.1-sol at $2 and $10, which is the number to sanity-check any contractor’s monthly API ask against. And the batch lane halves everything at all three providers without negotiation: OpenAI’s batch table posts gpt-6.1-sol at $1.00 and $5.00, Anthropic’s batch tokens for Sonnet 5.5 list $1.00 and $5.00, and Google documents its Batch API at a 50 percent cost reduction on the same page.
The Output Ratio: Why Every Quote Needs Two Price Columns
The ratio that shapes AI API pricing more than anything else on the page: output is always the expensive half. The ratio holds at every provider: Luna charges five times its input rate for output, Fable 5.1 and Astra charge five times at $50 against $10, and Sonnet 5.5 charges five times at $10 against $2. The engineering meaning is direct. Two calls that produce the same token count can differ in cost by a factor of six purely on how much they generate, so trimming output is six times more valuable than trimming an echo in the prompt. When you quote a client, price the bill as two columns, input and output, and the estimate survives a provider switch.
The rate card moves, and the dates are public. AI API pricing is a calendar as much as a table. Google’s increases are printed on its pricing page with effective dates. Carbon and Argon double on December 1, 2026. The Flash family doubles on January 1, 2027. Anthropic’s batch discounts and OpenAI’s cached-input rates have already reset the practical floor twice this year, and our running tally of the decline in AI inference cost shows serving prices down roughly tenfold across two years. Set a calendar reminder against the three pricing pages above; that costs nothing and protects every quote you have out.
What the Rates Mean for Freelancers Quoting AI Builds
A build’s maintenance promise is a price-list promise. When you commit to hosting a client’s AI feature for a fixed monthly fee, the small-model tier is the only tier whose rates make that fee stable: Haiku 5.5 at $0.75 for a 10-million-token month leaves headroom that survives any provider’s next increase. Quoting the frontier tier into a fixed-price retainer is a different business decision, one you should tie to a rate-adjustment clause, because Google has already scheduled a doubling into the contract window. The model-choice work we described in 51Talk-style comparisons applies here too: pick the tier that fits the task, then check the rate card twice before signing.
FAQ: AI API Pricing
Which provider has the cheapest AI API pricing in 2026?
Depends on the tier, and the answer is verified on the pages themselves. The cheapest small model on the Big Three’s own rate cards is Anthropic’s Claude Haiku 5.5 at $0.05 per million input tokens for prompts up to 100,000 tokens. The cheapest frontier model this month is Google’s Gemini 4 Carbon at $2 per million input, and it doubles on December 1, 2026. AI API pricing is tier-specific, so read the tier you will actually run at production volume before you compare providers at all.
What does 1 million tokens cost?
One million tokens runs from $0.15 input and $0.50 output on OpenAI’s smallest GPT-6 model to $10 input and $50 output on the OpenAI and Anthropic frontier models. Concretely, a million tokens is roughly 750,000 words of input, so the frontier rate is about $0.013 per thousand words, while the smallest models price the same thousand words at a fifth of a cent.
Is the batch lane really 50 percent off everywhere?
Yes, and it is documented on all three rate cards in November 2026: OpenAI’s batch table halves its standard rates (gpt-6.1-sol at $1.00 input, $5.00 output), Anthropic’s batch tokens for Sonnet 5.5 read $1.00 and $5.00, and Google states its Batch API at a 50 percent cost reduction. The tradeoff is latency and scheduling, not price.
When do Google’s AI API prices increase?
Two dates, posted on the pricing page itself. Gemini 4 Carbon and Gemini 4 Argon double from $2/$10 to $4/$20 per million tokens on December 1, 2026. Gemini 3.8 Flash and 3.6 Flash double from $0.75/$3.75 to $1.50/$7.50 on January 1, 2027. Both increases were live on the page when this article pulled its rates in mid-November 2026.
Why do output tokens cost more than input tokens?
Generation is the expensive computation. A model reads input tokens in one pass and produces output tokens one at a time, each step paying full inference cost, which is why the posted ratio runs 5:1 up to 10:1 across the tiers. Practically, it means the cheapest way to cut an API bill is to ask for shorter answers, and the second cheapest is prompt caching, where available.
Do cached input prices apply automatically?
OpenAI and Google price cached repeats automatically when request prefixes repeat, at $0.01 per million on Luna and $0.10 on Carbon’s implicit caching. Anthropic’s Sonnet 5.5 cache hits match at $0.10, but its cache writes carry their own 5-minute and 1-hour rates ($2.50 and $4.00 per million on Sonnet), so a cache-heavy design prices differently at Anthropic than it does at OpenAI. We compared the three caching systems in detail on the cache convergence page.
Financial Disclaimer
This article compares published model rates and computes illustrative costs for budgeting only. Rates change without notice; verify every figure against the provider’s own pricing page before making a purchase decision. Nothing here is financial or investment advice.
Sources and Further Reading
Rates were pulled from each provider’s own pricing documentation in November 2026: OpenAI API pricing, Anthropic Claude pricing, and Google Gemini API pricing. For the mechanics underneath every rate, read AI inference cost and the cache convergence analysis. For the decision these prices feed, see fine-tuning vs RAG vs prompting.
Cite this pagePublished Nov 16, 2026
Agron, E. (2026, November 16). AI API Pricing Compared: What the Big Three Charge Per Million Tokens. WorldNgayon. https://worldngayon.com/ai-api-pricing-compared/
Agron, Edmon. “AI API Pricing Compared: What the Big Three Charge Per Million Tokens.” WorldNgayon, 16 Nov. 2026, worldngayon.com/ai-api-pricing-compared/.
Agron, Edmon. “AI API Pricing Compared: What the Big Three Charge Per Million Tokens.” WorldNgayon, November 16, 2026. https://worldngayon.com/ai-api-pricing-compared/.
Agron, E. (2026) ‘AI API Pricing Compared: What the Big Three Charge Per Million Tokens’, WorldNgayon, 16 November. Available at: https://worldngayon.com/ai-api-pricing-compared/.
E. Agron, “AI API Pricing Compared: What the Big Three Charge Per Million Tokens,” WorldNgayon, Nov. 16, 2026. [Online]. Available: https://worldngayon.com/ai-api-pricing-compared/
@online{agron2026aiapi,
author = {Agron, Edmon},
title = {{AI API Pricing Compared: What the Big Three Charge Per Million Tokens}},
year = {2026},
date = {2026-11-16},
organization = {WorldNgayon},
url = {https://worldngayon.com/ai-api-pricing-compared/},
urldate = {}
}TY - ELEC AU - Agron, Edmon TI - AI API Pricing Compared: What the Big Three Charge Per Million Tokens T2 - WorldNgayon PY - 2026 DA - 2026/11/16/ UR - https://worldngayon.com/ai-api-pricing-compared/ Y2 - ER -
Cite the version you read; the page may be updated. Zotero and Mendeley can also import this page directly.






