Unofficial guide. Not affiliated with Meta.

Last updated:

AI model API pricing, compared

Every model we track, at list price per 1M tokens, and what the same coding workload costs on each. Prices are checked on the vendor pages.

API cost per task, model by model

Pick a codebase size and how many agent tasks you run a month. The table prices the same workload on every model at list price.

ModelInput / output per 1MPer taskPer month

Same assumptions as the calculator (60% of input from cache). Newer models often use fewer tokens per task, so treat this as a like-for-like list-price comparison.

Price list

ModelReleasedInputCache hitOutputContextSource
Claude Sonnet 5.5 AnthropicSep 28, 2026$2$0.20$101MOfficial
Claude Opus 5.5 AnthropicSep 22, 2026$4$0.20$201MOfficial
Claude Fable 5.1 AnthropicSep 1, 2026$10$0.25$50—Official
Claude Haiku 4.5 AnthropicOct 15, 2025$1$0.10$5200KOfficial
GPT-6.1 Sol OpenAISep 29, 2026$2$0.10$10—Official
GPT-6 Sol OpenAISep 22, 2026$2$0.60$10—Official
GPT-6 Luna OpenAISep 22, 2026$0.10$0.01$0.50—Official
GPT-6 Astra OpenAISep 4, 2026$10$1$501.05MOfficial

Which model for what

Announced or rumored, not available

Claude Sonnet 5.5 · Claude Opus 5.5 · Sonnet 5.5 vs Opus 5.5 · GPT-6 Sol · GPT-6 Astra · Opus 5.5 vs GPT-6 Astra · Opus 5.5 vs Fable 5.1 · Cost calculator

FAQ

What is Claude 5.5?

Not a single model. "Claude 5.5" is the family released in September 2026: Opus 5.5 (2026-09-22) and Sonnet 5.5 (2026-09-28), with Haiku 5.5 announced but not out yet.

Which AI model is cheapest for coding?

Per token, GPT-6 Luna and Claude Haiku 4.5. For agent coding quality per dollar, Sonnet 5.5 and GPT-6.1 Sol at $2 / $10 are the usual picks. Use the table above with your own task count.

Is GPT-6.1 Sol cheaper than GPT-6 Sol?

Same input and output price ($2 / $10), but cheaper cache hits: $0.10 per 1M against $0.60.

Independent site, not affiliated with Anthropic, OpenAI or Meta. USD list prices before tax.