Pricing index · Last verified 2026-08-23

LLM API Pricing Comparison: OpenAI, Claude & Gemini

Short answer: standard text rates run $0.20–$5.00 per 1M input tokens and $1.20–$25.00 per 1M output tokens across the major hosted models. The full rate card, a workload cost calculator, and every official source — verified weekly — are below.

Rate card

LLM API pricing by model

Standard on-demand text-token pricing per 1,000,000 tokens, with context window and output caps. Caching, batch and long-context rules can change the real bill — see the notes.

ModelInput / 1MOutput / 1MContext windowMax outputSource
GPT-5.6 Sol
OpenAI
$4.00$20.001,050,000128,000OpenAI
GPT-5.6 Terra
OpenAI
$2.00$12.001,050,000128,000OpenAI
GPT-5.6 LunaLowest
OpenAI
$0.20$1.201,050,000128,000OpenAI
Claude Opus 5
Anthropic
$5.00$25.001,000,000128,000Anthropic
Claude Sonnet 5
Anthropic
$2.00 *$10.00 *1,000,000128,000Anthropic
Claude Haiku 4.5
Anthropic
$1.00$5.00200,00064,000Anthropic
Gemini 3.6 Flash
Google
$0.75$3.751,048,57665,536Google
Gemini 3.5 Flash-Lite
Google
$0.30$2.501,048,57665,536Google

* Claude Sonnet 5's introductory $2.00/$10.00 pricing is now the standard rate — the planned Sep 1, 2026 increase to $3.00/$15.00 was cancelled. Gemini 3.6 Flash promotional $0.75/$3.75 rates run through Dec 31, 2026, then move to $1.50/$7.50. OpenAI GPT-5.6 models bill higher long-context rates on prompts over 272K input tokens (2x input, 1.5x output). Prices are informational estimates in USD per provider-published token rates and may exclude caching, batch discounts, tools, regional premiums or taxes. The provider invoice is the source of truth.

Live calculator

What would your workload cost per month?

Pick a typical workload preset or enter your own per-request token volumes and daily request count. Costs use the standard rates above — no caching or batch discounts applied.

Presets:
Cheapest 30-day total
Most expensive 30-day total
ModelPer requestPer day30-day estimate
Standard rates: text token pricing. Excluded: caching, batch, tools and taxes. Pricing data verified —
Worked example

A support chatbot at 5,000 conversations a day

Assume 2,000 input tokens and 500 output tokens per conversation — a typical support-bot shape. Monthly volumes: 300M input + 75M output tokens. Here is the 30-day bill at standard rates:

ModelCost per conversation30-day estimate
GPT-5.6 Sol
OpenAI
$0.0180$2,700.00
GPT-5.6 Terra
OpenAI
$0.0100$1,500.00
GPT-5.6 LunaLowest
OpenAI
$0.0010$150.00
Claude Opus 5
Anthropic
$0.0225$3,375.00
Claude Sonnet 5
Anthropic
$0.0090$1,350.00
Claude Haiku 4.5
Anthropic
$0.0045$675.00
Gemini 3.6 Flash
Google
$0.0034$506.25
Gemini 3.5 Flash-Lite
Google
$0.0019$277.50

Same workload, 18x price spread. Model quality, latency and context needs still apply — "cheapest" here is a pure arithmetic result at standard rates, not a recommendation.

How to read this

Three rules before you compare LLM API prices.

1. Split input and output. Output tokens cost 5–8x more than input tokens across every provider, because generation is the expensive part of inference. A workload that is 90% output (long-form generation) ranks models completely differently from one that is 90% input (classification, extraction).

2. Watch the surcharges. OpenAI's GPT-5.6 family bills 2x input and 1.5x output once a prompt exceeds 272K input tokens. Long-document workloads can silently double in cost — check fit first with the Context Window Checker.

3. Standard rates are the ceiling, not the floor. Prompt caching, batch APIs and committed-use tiers routinely cut real bills 30–80%. Use this page to shortlist models, then confirm discount mechanics on the provider's pricing page.

Deep dive: OpenAIModel-by-model rates, long-context rules and workload costs on the OpenAI API Pricing page.
Deep dive: AnthropicOpus 5, Sonnet 5 and Haiku 4.5 rates plus the cancelled Sonnet price increase: Claude API Pricing.
Deep dive: Google3.6 Flash promotional window and Flash-Lite budget rates: Gemini API Pricing.
Compare: OpenAI vs ClaudeSame workload billed both ways, break-even volumes and the 272K rule: OpenAI vs Claude Pricing.
Compare: Claude vs GeminiPromo-window math and the Jan 1, 2027 flip: Claude vs Gemini Pricing.
Compare: OpenAI vs GeminiTwo expiring promos, the 272K long-context flip and the output-token ceiling: OpenAI vs Gemini Pricing.
Guide: words to tokensWhat 1,000 words costs by content type and model: How Many Tokens in 1,000 Words?.
Per-million viewWhat a flat 1M tokens costs across all three providers: How Much Does 1 Million AI Tokens Cost?
Budget your own workloadModel monthly and annual spend with the API Cost Calculator, or measure a real prompt with the Token Counter.
FAQ

LLM API pricing questions

Which LLM API is the cheapest?

At standard rates among the tracked models, GPT-5.6 Luna is the cheapest at $0.20 per 1M input and $1.20 per 1M output tokens, followed by Gemini 3.5 Flash-Lite at $0.30/$2.50. The cheapest model for your workload depends on your input-to-output ratio — use the calculator on this page.

How much does the OpenAI API cost compared to Claude?

At flagship level, GPT-5.6 Sol ($4.00/$20.00 per 1M input/output, promotional through at least Nov 21, 2026) now undercuts Claude Opus 5 ($5.00/$25.00) on both input and output. At mid tier, GPT-5.6 Terra ($2.00/$12.00) matches Claude Sonnet 5's $2.00/$10.00. At the budget tier, GPT-5.6 Luna ($0.20/$1.20) undercuts Claude Haiku 4.5 ($1.00/$5.00) by 5x on input.

Why do output tokens cost more than input tokens?

Input tokens are processed in parallel in a single pass, while output tokens are generated one at a time, each requiring a full model forward pass. Providers price that asymmetry directly: across the tracked models, output rates run 5–8x higher than input rates.

Are these the real prices I will pay?

These are standard on-demand text-token rates from official provider pricing pages, re-verified weekly. Real bills are often lower: prompt caching, batch APIs and volume tiers can cut costs 50% or more, while long-context surcharges (for example OpenAI prompts over 272K input tokens) can raise them. The provider invoice is always the source of truth.

How often does LLM API pricing change?

Frequently, and mostly downward. Providers adjust rates and model lineups every few months — for example, OpenAI cut GPT-5.6 Sol by over 20% to $4.00/$20.00 on Aug 21, 2026 (promotional through at least Nov 21, 2026), Anthropic made Claude Sonnet 5's introductory $2.00/$10.00 pricing permanent instead of raising it, and Google is running Gemini 3.6 Flash at promotional $0.75/$3.75 rates through Dec 31, 2026. We re-verify every rate on this page against official sources weekly; the Last verified date shows the latest check.

Methodology

Where these numbers come from.

All rates are taken from official provider pricing pages and re-verified weekly; the "Last verified" date reflects the latest full check. The calculator multiplies your token volumes by published per-million-token rates — it does not estimate tokenization and does not apply caching, batch or tiered discounts unless noted.

Official sources: OpenAI model docs, Anthropic pricing, Google Gemini API pricing. Always confirm the provider invoice before making purchasing decisions.

Next step: measure itPaste your real prompt into the AI Token Counter.
Next step: budget itTurn rates into monthly spend with the API Cost Calculator.
Next step: convert itTranslate token budgets into word counts with the Tokens ↔ Words converter.
What 100K tokens is~75,000 words, 150 pages, and the chunk size that stays under OpenAI's 272K surcharge: How Many Words Is 100K Tokens?.
Context windows explainedInput and output share one budget — window sizes, output caps and what it costs to fill one: What Is a Context Window?.
What a token isDefinition, conversion by content type, and why output costs 5–8x input: What Is an AI Token?
HTML token savingsCleaning scraped markup removes 47–68% of input tokens: HTML Token Savings.
1M tokens in words750,000 words, 1,500 pages — and why only 5 of 8 models accept it in one request: How Many Words Is 1M Tokens?
Trim the promptFind repeated instructions, markup and filler: Prompt Weight Analyzer
How we verifySources, weekly cadence and what our estimates exclude: Methodology · Pricing changelog.