As of 13 October 2026, Anthropic and OpenAI list their models at the same three price points: $10 per million input tokens and $50 per million output tokens for Claude Fable 5.1 and gpt-6-astra, $2 and $10 for Claude Sonnet 5.5 and gpt-6.1-sol, and $0.10 and $0.50 for Claude Haiku 5.5 and gpt-6-luna. Google's three Gemini models in this comparison list at or below the middle price point. Those are list prices. The bill you actually pay depends on tokens per task, caching, batch use and prompt length, which is where most budget mistakes happen. Every price below was read from the provider's own pricing page on 13 October 2026.
Why prices moved this month
Several launches in five weeks changed the table. OpenAI's changelog shows GPT-6 Astra released on 3 September, GPT-6 Sol and Luna on 22 September, and GPT-6.1 Sol on 29 September. Anthropic released Claude Haiku 5.5 on 7 October. Google's pricing page was last updated on 9 October and lists gemini-3.8-flash and a preview Pro model, gemini-3.1-pro-preview.
The prices, side by side
Standard-tier prices in US dollars per million tokens. Prompt-length thresholds differ by provider, so they get their own column.
| Model | Input / 1M | Output / 1M | Cached input / 1M | Long prompts (input / output) |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $0.25 | Standard price up to 1M tokens |
| Claude Opus 5.5 | $4 | $20 | $0.20 | Standard price up to 1M tokens |
| Claude Sonnet 5.5 | $2 | $10 | $0.10 | Standard price up to 1M tokens |
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.01 | Over 100,000 tokens: $0.50 / $2.50 |
| gpt-6-astra | $10 | $50 | $1.00 | Over 272,000 tokens: $20 / $75 |
| gpt-6.1-sol | $2 | $10 | $0.10 | Over 272,000 tokens: $4 / $15 |
| gpt-6-luna | $0.10 | $0.50 | $0.01 | Over 272,000 tokens: $0.20 / $0.75 |
| gemini-3.1-pro-preview | $2 | $12 | See Google's page | Over 200,000 tokens: $4 / $18 |
| gemini-3.8-flash | $0.75 | $3.75 | See Google's page | $1.50 / $7.50 from 1 January 2027 |
| gemini-3.5-flash-lite | $0.30 | $2.50 | See Google's page | None listed |
All three providers discount asynchronous batch work by 50%: Anthropic's Batch API, OpenAI's Batch and Flex, and Google's Batch API. Our token cost calculator uses these prices for nine of the ten models (it leaves out Claude Opus 5.5) and works out a per-request and per-day cost for your own prompt.
What the Haiku 5.5 price cut really means
Anthropic prices Claude Haiku 5.5 at $0.10 and $0.50 per million tokens for prompts up to 100,000 tokens, against $1 and $5 for Claude Haiku 4.5. That is 90% lower per token. Anthropic's own announcement says the model costs around 75% less to run on average, and explains the gap: Haiku 5.5 has an updated tokenizer and uses slightly more tokens per task. Its documentation puts a number on the tokenizer, saying the same text counts as about 30% more tokens than on Haiku 4.5.
Our arithmetic from those two figures: 30% more tokens at 10% of the price is about 13% of the old bill for the same text, a saving of roughly 87%. Anthropic's stated average is lower, which suggests other differences between the models also matter. The practical point is that a per-token price cut and a per-task saving are different numbers. Budget with a cost per task measured on your own prompts, not with the headline percentage.
Anthropic positions Haiku 5.5 for high-volume, latency-sensitive work such as classification, extraction and routing, and as a subagent. OpenAI's gpt-6-luna lists at exactly the same input and output prices.
Same list price, different bill
- Caching. Anthropic charges cache reads at 10% of the base input price (5% on Claude Opus 5.5 and Sonnet 5.5, 2.5% on Claude Fable 5.1). OpenAI lists cached input at $1.00 for gpt-6-astra, $0.10 for gpt-6.1-sol, $0.20 for gpt-6-sol and $0.01 for gpt-6-luna, so two models with the same $2 and $10 list price differ on cached input. Writing to a cache costs more than ordinary input, 1.25 times the input price for the shorter Anthropic cache and for OpenAI.
- Prompt length. Claude Haiku 5.5 charges more once a prompt passes 100,000 tokens ($0.50 and $2.50). Gemini 3.1 Pro passes to $4 and $18 above 200,000 tokens. OpenAI's GPT-6 models step up above 272,000 tokens. Anthropic says its other current models include the full 1M-token context at the standard price.
- Where the request runs. Anthropic adds a 10% multiplier when you pin inference to the US on Claude 4.6 and later models. OpenAI charges a 10% uplift on regional processing endpoints for models released on or after 5 March 2026.
- Speed. Both Anthropic and OpenAI sell faster modes at a premium. OpenAI's Fast mode lists at twice the standard rate.
- Thinking. Google states that its output price includes thinking tokens. Anthropic's models page lists a default effort level for each model and says adaptive thinking decides how much to think, steered by effort. Either way, thinking depth is a cost lever worth testing on your own workload.
Prices change, and some already carry a date
- Google lists
gemini-3.8-flashat $0.75 and $3.75 through 31 December 2026, and $1.50 and $7.50 from 1 January 2027. - Anthropic's pricing page says the $2 and $10 introductory price of Claude Sonnet 5 is now its standard price, and that the increase to $3 and $15 planned for 1 September 2026 will not happen.
- OpenAI's changelog notes promotional pricing on GPT-5.6 Sol, an older model, through at least 21 November 2026. We found no promotional or expiry note for the GPT-6 models.
We wrote about what to do when a vendor reprices overnight in what happens when your AI vendor changes their pricing. The short version: keep model names in configuration, not scattered through code, and leave headroom in the budget.
A worked example
Take a support agent that handles 10,000 conversations a month. Each conversation makes 6 model calls, and each call sends 3,000 input tokens and returns 400 output tokens. That is 180 million input tokens and 24 million output tokens a month. At list price, with no caching or batch discount:
| Model | Monthly cost at list price |
|---|---|
| Claude Fable 5.1 | $3,000 |
| Claude Opus 5.5 | $1,200 |
| Claude Sonnet 5.5 | $600 |
| Claude Haiku 5.5 | $30 |
| gpt-6-astra | $3,000 |
| gpt-6.1-sol | $600 |
| gpt-6-luna | $30 |
| gemini-3.1-pro-preview | $648 |
| gemini-3.8-flash | $225 ($450 from 1 January 2027) |
| gemini-3.5-flash-lite | $114 |
The spread from top to bottom is 100 times ($3,000 against $30) at both Anthropic and OpenAI. Treat the numbers as arithmetic, not as a forecast. They give every model the same token counts, which is not true because tokenizers differ, and they say nothing about quality. They also show why routing matters: a model that costs a hundredth as much only saves money on the calls where it is good enough.
How we would use this
This is our approach, not a rule. Route work by task instead of picking one model for everything: use the smallest model that passes your own evaluation for classification, extraction and routing, and spend on a stronger model where a mistake is expensive. Measure cost per completed task, since that is the number that survives tokenizer changes. Keep model names configurable, and re-check prices monthly. For the method behind agent budgets, see how to estimate the cost of an AI agent, choosing a reasoning effort budget and how to choose between Claude, GPT and Gemini.
How we approach this
We build AI systems that are not tied to one model vendor, which is why we track prices across all of them, and we say plainly when a cheaper model would not be good enough for the job.
Sources, all read on 13 October 2026: Anthropic's pricing page, Anthropic's Claude Haiku 5.5 announcement, Claude Haiku 5.5 documentation, OpenAI's pricing page, OpenAI's API changelog and Google's Gemini API pricing page.