AI Receipts

Tracker · Last checked

AI model price tracker

What the big AI APIs cost, in US dollars per million tokens (a token is roughly three-quarters of a word). Every row is copied from the provider's official pricing page, with the link and the date we checked it.

Standard tier, text, shortest context bracket. Excludes batch, caching and fast-mode pricing. PROMO = temporary price with a scheduled increase.
ModelInputOutputNotesSource · checked
Anthropic
Claude Opus 5.5		$4.00$20.00Batch API is 50% off. Fast mode costs $8 / $40.platform.claude.com
2026-10-14
Anthropic
Claude Sonnet 5.5		$2.00$10.00—platform.claude.com
2026-10-14
Anthropic
Claude Haiku 5.5		$0.10$0.50Prompts up to 100K tokens. Over 100K: $0.50 / $2.50.platform.claude.com
2026-10-14
Google
Gemini 4 ArgonPROMO		$2.00$10.00Introductory price through Nov 30, 2026. $4 / $20 starting Dec 1, 2026. Output includes thinking tokens.ai.google.dev
2026-10-14
Google
Gemini 3.8 FlashPROMO		$0.75$3.75Price through Dec 31, 2026. $1.50 / $7.50 starting Jan 1, 2027.ai.google.dev
2026-10-14
OpenAI
GPT-6 Astra		$10.00$50.00Up to 272K input tokens. Longer context: $20 / $75.platform.openai.com
2026-10-14
OpenAI
GPT-6.1 Sol		$2.00$10.00Up to 272K input tokens. Longer context: $4 / $15.platform.openai.com
2026-10-14
OpenAI
GPT-6 Luna		$0.10$0.50Up to 272K input tokens. Longer context: $0.20 / $0.75.platform.openai.com
2026-10-14

How to read this

Per-token price is only half the bill. A model that is cheaper per token can still cost more per task if it writes more tokens to get there. Our first issue found exactly that: Gemini 4 Argon is cheaper than GPT-6 Astra per task only while its introductory price lasts.

We only list a price after reading it on the provider's own pricing page. When a price changes we update the row and the date. Spot something stale? See our corrections policy. Raw data: pricing.json (CC BY 4.0, please link back).