# LLM API prices for agents, tracked

> Current list prices per million tokens for Claude, GPT and Gemini models, with the caching and batch discounts that decide what an agent costs, and the promotional prices that end soon. Checked monthly.

- URL: https://bardak.ai/posts/llm-api-prices
- Author: Ivan Bardak (https://bardak.ai/pages/about)
- Published: 2026-11-10
- Updated: 2026-11-16

List prices for language model APIs change every few months, some of them are promotional and end on a fixed date, and [the discounts that matter most for agents](https://bardak.ai/posts/prompt-caching-batching-routing), caching and batch processing, work differently at each provider. This tracker keeps the current prices of the main models from Anthropic, Google and OpenAI in one place, with the conditions attached to them. I check it every month against the providers' own pricing pages.

**As of 1 November 2026:** the mid-range models from all three providers cost $2 per million input tokens and $10 to $12 per million output tokens, the smaller models cost $0.10 to $1 per million input tokens, and two Google prices are promotional: Gemini 4 Argon doubles on 1 December 2026, Gemini 3.8 Flash and 3.6 Flash on 1 January 2027.

Price per token is not what an agent costs. An agent calls the model many times per task, and failed attempts cost money too. How to turn these prices into a cost per completed task is in [What an AI agent really costs](https://bardak.ai/posts/ai-agent-cost-per-task); the calculator at the end of this page does the arithmetic.

## LLM API prices at a glance

**LLM API list prices per million tokens**

| Model | In $/1M | Out $/1M |
| --- | --- | --- |
| Claude Fable 5.1 | 10.00 | 50.00 |
| Claude Opus 5.5 | 4.00 | 20.00 |
| Claude Sonnet 5.5 | 2.00 | 10.00 |
| Claude Haiku 5.5 | 0.10 | 0.50 |
| GPT-6 Astra | 10.00 | 50.00 |
| GPT-6.1 Sol | 2.00 | 10.00 |
| GPT-6.1 Terra | 1.00 | 5.00 |
| GPT-6 Luna | 0.10 | 0.50 |
| Gemini 4 Argon | 2.00 | 10.00 |
| Gemini 3.1 Pro Preview | 2.00 | 12.00 |
| Gemini 3.8 Flash | 0.75 | 3.75 |
| Gemini 3.5 Flash-Lite | 0.30 | 2.50 |
| Gemini 3.1 Flash-Lite | 0.25 | 1.50 |

Claude models are from Anthropic, GPT from OpenAI and Gemini from Google. Standard (pay-as-you-go) prices in US dollars per million tokens, before caching and batch discounts, for prompts of ordinary length. Output prices include reasoning tokens at all three providers. The Gemini 4 Argon and 3.8 Flash prices are promotional. Source: Anthropic, OpenAI and Google pricing pages, accessed 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/llm-api-list-prices-per-million-tokens.csv

## Claude API pricing

Anthropic's current models run from Claude Haiku 5.5 at $0.10 and $0.50 per million input and output tokens to Claude Fable 5.1 at $10 and $50. Where they differ most is reading from the cache: a fortieth of the input price on Fable 5.1, a twentieth on Opus 5.5 and Sonnet 5.5, a tenth on the others. The Batch API halves every price, and Claude models from 4.6 on keep the same price across their whole 1M-token context, except Haiku 5.5: a prompt over 100,000 tokens costs five times as much ([Anthropic](https://platform.claude.com/docs/en/about-claude/pricing)).

**Claude API prices per million tokens**

| Model | Input | Output | Cache read |
| --- | --- | --- | --- |
| Claude Fable 5.1 | 10.00 | 50.00 | 0.25 |
| Claude Opus 5.5 | 4.00 | 20.00 | 0.20 |
| Claude Sonnet 5.5 | 2.00 | 10.00 | 0.10 |
| Claude Haiku 5.5 | 0.10 | 0.50 | 0.01 |
| Claude Opus 5 | 5.00 | 25.00 | 0.50 |
| Claude Sonnet 5 | 2.00 | 10.00 | 0.20 |
| Claude Haiku 4.5 | 1.00 | 5.00 | 0.10 |

US dollars per million tokens. The Batch API is half price on input and output. Writing to the cache costs 1.25 times the input price for 5 minutes and twice the input price for an hour. Haiku 5.5 prices are for prompts up to 100,000 tokens. Claude Mythos 5.1 costs the same as Fable 5.1 but is open only to verified organisations. Source: Anthropic pricing and prompt caching pages, accessed 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/claude-api-prices-per-million-tokens.csv

## OpenAI API pricing

OpenAI's GPT-6 family spans GPT-6 Luna at $0.10 and $0.50 to GPT-6 Astra at $10 and $50, with GPT-6.1 Sol and GPT-6.1 Terra in between. Batch and Flex processing halve the price, Fast mode (called Priority until July 2026) doubles it for the GPT-6 family, and a prompt over 272,000 tokens pays twice the input price and one and a half times the output price. Cached input costs a tenth of the input price, and a twentieth on GPT-6.1 Sol and Terra ([OpenAI](https://developers.openai.com/api/docs/pricing)).

**OpenAI API prices per million tokens**

| Model | Input | Output | Cached in |
| --- | --- | --- | --- |
| GPT-6 Astra | 10.00 | 50.00 | 1.00 |
| GPT-6.1 Sol | 2.00 | 10.00 | 0.10 |
| GPT-6.1 Terra | 1.00 | 5.00 | 0.05 |
| GPT-6 Luna | 0.10 | 0.50 | 0.01 |

US dollars per million tokens, Standard tier, prompts up to 272,000 tokens. Batch and Flex are half price. Writing to the cache costs 1.25 times the input price for 30 minutes and 2.5 times for 12 hours (Responses API only). Regional and FedRAMP endpoints add 10%. Older models, from GPT-6 Sol back to GPT-4.1 and the o-series, keep their own prices on OpenAI's page. Source: OpenAI pricing and prompt caching pages, accessed 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/openai-api-prices-per-million-tokens.csv

## Gemini API pricing

Google's line-up carries the two promotional prices with a fixed end date: Gemini 4 Argon at $2 and $10 until 30 November, and Gemini 3.8 Flash and 3.6 Flash at $0.75 and $3.75 until 31 December. Reading from the cache costs a tenth of the input price (a twentieth on Gemini 4 Argon), explicit caches also pay for storage by the hour, and Gemini 4 Argon caches only implicitly. Batch halves the price; the Priority tier costs 1.8 times the standard one ([Google](https://ai.google.dev/gemini-api/docs/pricing)).

**Gemini API prices per million tokens**

| Model | Input | Output | Cache read |
| --- | --- | --- | --- |
| Gemini 4 Argon | 2.00 | 10.00 | 0.10 |
| Gemini 3.1 Pro Preview | 2.00 | 12.00 | 0.20 |
| Gemini 3.8 Flash | 0.75 | 3.75 | 0.075 |
| Gemini 3.5 Flash-Lite | 0.30 | 2.50 | 0.03 |
| Gemini 3.1 Flash-Lite | 0.25 | 1.50 | 0.025 |

US dollars per million tokens, standard paid tier; output includes thinking tokens. Batch is half price. Gemini 4 Argon and 3.8 Flash are at promotional prices that double on 1 December 2026 and 1 January 2027; Gemini 3.6 Flash costs the same as 3.8 Flash. Gemini 3.1 Pro Preview prices are for prompts up to 200,000 tokens; above that, $4 and $18. Source: Google Gemini API pricing page, accessed 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/gemini-api-prices-per-million-tokens.csv

## Prices that change soon

- **Gemini 4 Argon:** $2 and $10 until 30 November 2026; from 1 December it is $4 and $20, and cache reads go from $0.10 to $0.20 ([Google](https://ai.google.dev/gemini-api/docs/pricing)).
- **Gemini 3.8 Flash and 3.6 Flash:** $0.75 and $3.75 until 31 December 2026; from 1 January 2027 they are $1.50 and $7.50, and cache reads and storage double too ([Google](https://ai.google.dev/gemini-api/docs/pricing)).
- **GPT-5.6 Sol:** OpenAI says its promotional price of $4 and $20 runs at least through 21 November 2026, without saying what follows ([OpenAI](https://developers.openai.com/api/docs/pricing)).
- **Gemini 3.1 Pro Preview** is still a preview, so check its price again before you build a cost estimate on it.

If a cost estimate for your agent uses one of these models, run it again at the later price.

## What one agent call costs

Agents send far more input than output: the instructions, the tool definitions and the steps so far travel with every call. Here is the list price of 1,000 calls of 10,000 input tokens and 1,000 output tokens each, a typical shape for one step of an agent. My arithmetic, without caching:

**Cost of 1,000 agent calls at list price**

| model | Input tokens | Output tokens |
| --- | --- | --- |
| Claude Fable 5.1 | 100.00 | 50.00 |
| GPT-6 Astra | 100.00 | 50.00 |
| Claude Opus 5.5 | 40.00 | 20.00 |
| Gemini 3.1 Pro Preview | 20.00 | 12.00 |
| Claude Sonnet 5.5 | 20.00 | 10.00 |
| GPT-6.1 Sol | 20.00 | 10.00 |
| Gemini 4 Argon | 20.00 | 10.00 |
| GPT-6.1 Terra | 10.00 | 5.00 |
| Gemini 3.8 Flash | 7.50 | 3.75 |
| Gemini 3.5 Flash-Lite | 3.00 | 2.50 |
| Gemini 3.1 Flash-Lite | 2.50 | 1.50 |
| Claude Haiku 5.5 | 1.00 | 0.50 |
| GPT-6 Luna | 1.00 | 0.50 |

Each call: 10,000 input and 1,000 output tokens at the standard prices above. Input is 55% to 67% of the bill for every model here: output costs five to about eight times more per token, but each call sends ten times more input. Gemini 4 Argon goes to $60 and Gemini 3.8 Flash to $22.50 when their promotional prices end. Source: My arithmetic from the providers' pricing pages, 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/cost-of-1-000-agent-calls-at-list-price.csv

A hundredfold separates the cheapest and the dearest model for the same call. Whether the cheap one is good enough is a question for your own evaluation set, not for this table: the reliability numbers are in [AI agent reliability benchmarks, tracked](https://bardak.ai/posts/ai-agent-reliability-benchmarks).

## Caching and batch discounts

For agents, the discounts often matter more than the list price, because most of each call repeats the call before it. All three providers halve the price for batch work. Caching differs:

**Caching and batch discounts by provider**

| Provider | Reading from cache | Writing to cache | Batch | Smallest cacheable prompt |
| --- | --- | --- | --- | --- |
| Anthropic | A tenth of the input price; a twentieth on Opus 5.5 and Sonnet 5.5; a fortieth on Fable 5.1 | 1.25 times the input price for 5 minutes, twice for 1 hour | Half price on input and output; stacks with caching | 512 tokens on the current models |
| OpenAI | Automatic; a tenth of the input price, a twentieth on GPT-6.1 Sol and Terra | 1.25 times the input price for 30 minutes, 2.5 times for 12 hours (GPT-5.6 and later) | Half price with Batch or Flex | 1,024 tokens on GPT-5.6 and later |
| Google | Automatic (implicit); a tenth of the input price, a twentieth on Gemini 4 Argon, which caches only implicitly | Explicit caches pay storage per hour: $0.50 per million tokens on 3.8 Flash, $4.50 on 3.1 Pro Preview | Half price | 4,096 tokens on Gemini 4 Argon, 3.8 Flash and 3.1 Pro Preview |

Anthropic says caching pays off after a single cache read for the 5-minute duration. Source: Anthropic, OpenAI and Google pricing and caching documentation, accessed 1 November 2026. Data: https://bardak.ai/data/llm-api-prices/caching-and-batch-discounts-by-provider.csv

In practice: put the stable part of the prompt, instructions and tool definitions, at the start so it can be cached, and run evaluations and re-indexing as batches ([Anthropic](https://platform.claude.com/docs/en/build-with-claude/prompt-caching); [OpenAI](https://developers.openai.com/api/docs/guides/prompt-caching); [Google](https://ai.google.dev/gemini-api/docs/caching)).

## Why prices per token do not compare exactly

- **Tokenizers differ.** The same text becomes a different number of tokens at each provider, and even between model generations: Anthropic notes that its tokenizer from Claude 4.7 on produces about 30% more tokens for the same text ([Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing)).
- **Reasoning is billed as output.** Thinking tokens are charged at the output price at all three providers ([Anthropic](https://platform.claude.com/docs/en/build-with-claude/extended-thinking); [OpenAI](https://developers.openai.com/api/docs/guides/reasoning); [Google](https://ai.google.dev/gemini-api/docs/pricing)), and OpenAI notes a request may use from a few hundred to tens of thousands of them. A model that thinks longer costs more per task at the same list price.
- **Long prompts cost more.** Claude Haiku 5.5 is five times dearer above 100,000 prompt tokens, OpenAI doubles the input price above 272,000, and the Gemini 3.1 Pro Preview price holds only up to 200,000.

The result shows on the same evaluations: Artificial Analysis ran two models with the same $2 and $10 list prices, GPT-6.1 Sol and Claude Sonnet 5.5, at their highest reasoning settings, and paid $1.27 and $10.02 per task on its Intelligence Index, about eight times as much for the same list price ([Artificial Analysis](https://artificialanalysis.ai/models)). The full comparison is in [What an AI agent really costs](https://bardak.ai/posts/ai-agent-cost-per-task).

## Work out your own cost

Put a row from the price tables into the calculator, with the calls, tokens and success rate you measure on your own tasks:

_An interactive agent cost calculator stands here on the web page at https://bardak.ai/posts/llm-api-prices._

For a business case, the [AI business case template](https://bardak.ai/downloads/ai-business-case-template.xlsx) puts the same cost next to the value of one task and a break-even success rate.

## Changelog

- **1 November 2026.** Checked, no changes.
- **29 October 2026.** A section and a table per provider, with cache-read prices. Added GPT-6.1 Terra, Gemini 3.5 Flash-Lite and the older Claude models still on sale; cache reads at a fortieth of the input price on Claude Fable 5.1 and a twentieth on GPT-6.1 Sol and Terra; Google's caching prices and minimums; the GPT-5.6 Sol promotion.
- **27 October 2026.** First version: eleven models from Anthropic, OpenAI and Google, caching and batch discounts, two promotional prices with their end dates.

## Sources

- [Anthropic. Pricing.](https://platform.claude.com/docs/en/about-claude/pricing)
- [Anthropic. Prompt caching.](https://platform.claude.com/docs/en/build-with-claude/prompt-caching)
- [Anthropic. Extended thinking.](https://platform.claude.com/docs/en/build-with-claude/extended-thinking)
- [OpenAI. API pricing.](https://developers.openai.com/api/docs/pricing)
- [OpenAI. Prompt caching.](https://developers.openai.com/api/docs/guides/prompt-caching)
- [OpenAI. Reasoning models.](https://developers.openai.com/api/docs/guides/reasoning)
- [Google. Gemini API pricing.](https://ai.google.dev/gemini-api/docs/pricing)
- [Google. Context caching.](https://ai.google.dev/gemini-api/docs/caching)
- [Artificial Analysis. Models (Intelligence Index v5.0).](https://artificialanalysis.ai/models)
