Frontier AI API price tracker: what OpenAI, Anthropic and Google charge per million tokens

This is the running price tracker for frontier AI models on the three big developer platforms, built from each lab's own pricing page and dated October 14, 2026. If you are looking for the current price of an AI model API, the short version is that the flagship tier has settled at $10 per million input tokens and $50 per million output tokens at OpenAI and Anthropic, while Google just entered at a fifth of that with a price that doubles on December 1.

Prices below are US dollars per million tokens, standard tier, short context unless noted. Each line names the page we read. We will update this page whenever a lab changes a rate, so check the date before you quote a number.

The flagship tier

The highest-priced models sit at the same rate on both OpenAI and Anthropic:

• OpenAI GPT-6 Astra: $10 input, $50 output, $1 for cached input. Above 272,000 input tokens the whole request moves to $20 input and $75 output.
• Anthropic Claude Fable 5.1: $10 input, $50 output. Fable 5, the version before it, is listed at the same rate.

OpenAI also sells faster modes for Astra. Its Fast tier is $20 input and $100 output, and its Ultrafast tier is $60 input and $300 output. Batch use halves the standard rate to $5 and $25. OpenAI lists a 10 percent uplift for regional processing and FedRAMP endpoints.

The workhorse tier

This is where most production traffic lives, and prices have converged here too:

• Anthropic Claude Opus 5.5: $4 input, $20 output.
• Anthropic Claude Sonnet 5.5: $2 input, $10 output, with cache reads at $0.10.
• OpenAI GPT-6.1 Sol and GPT-6 Sol: $2 input, $10 output each, with cached input at $0.10 and $0.20 respectively.
• OpenAI GPT-5.6 Sol: $4 input, $20 output. GPT-5.6 Terra: $2 input, $12 output.
• Google Gemini 4 Argon, through November 30, 2026: $2 input, $10 output.
• Google Gemini 4 Argon from December 1, 2026: $4 input, $20 output.

Notice what the list says. At $2 and $10, three different labs now sell a model for the same price. Argon launched into that cluster on October 13, 2026, and Google's launch post states that cached input tokens are 95 percent cheaper than uncached ones. Google's pricing page says the model is not available on its free tier or its AI Pro plan, so it is a paid-API product.

The budget tier

The cheapest small models are an order of magnitude below the rest:

• OpenAI GPT-6 Luna: $0.10 input, $0.50 output.
• Anthropic Claude Haiku 5.5: $0.10 input and $0.50 output for prompts up to 100,000 tokens, then $0.50 and $2.50 above that. Its predecessor, Haiku 4.5, is still listed at $1 and $5.

We cover what changed in that release in our report on the Haiku 5.5 price and benchmarks. A fall from $1 to $0.10 in one generation is a tenfold cut, and it is the clearest price move in this table.

Google's tier table for Argon

Google publishes four paid tiers for Argon, each doubling on December 1:

• Standard: $2 and $10 now, $4 and $20 later.
• Batch and Flex: $1 and $5 now, $2 and $10 later.
• Priority: $3.60 and $18 now, $7.20 and $36 later.

Context caching is implicit only, because explicit caching is not supported for this model. Google Search grounding carries 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000 requests.

What list price hides

A per-token price is not a cost per task. A model that thinks longer bills more output tokens, and output is the expensive side at a typical 5 to 1 ratio on these lists. A cheaper model that needs two attempts can cost more than a dearer one that needs one. Cached input changes the picture for agents that resend long context, where a 90 to 95 percent cache discount does more than any headline cut.

Gizmodo's launch coverage compared Argon at $2 and $10 with GPT-6 Astra at $10 and $50. That comparison is accurate on list price, but two models that score differently are not interchangeable at the same budget, and none of the three labs publishes cost per solved task. Treat the table as the starting point for your own test.

Change log

• October 14, 2026: tracker opened. Argon general availability on October 13 added at the introductory rate, with the December 1 doubling recorded in advance.
• Next scheduled change: December 1, 2026, when every Argon tier doubles.

How to read the table

Compare like with like. The same dollar figure buys different amounts of work on different models, because tokenizers differ, so a million tokens is not a fixed amount of text across labs. Long-context surcharges apply to the whole request at OpenAI once input passes 272,000 tokens, while Anthropic's Haiku 5.5 steps up above 100,000 prompt tokens. If your prompts are long, read those footnotes first.

What we would do

For a new project, run your own task set on the workhorse tier first and only move up if the quality gap shows in your numbers. Budget for the December 1 change in Argon now, since it is already published. Check each lab's page before you commit, because these rates can change on a stated date, as Argon's will.