Claude Haiku 5.5 vs GPT-6 Luna: List Prices and Limits Side by Side

Claude Haiku 5.5 and GPT-6 Luna list the same base price: $0.10 per million input tokens and $0.50 per million output tokens. The two price lists split once prompts get long. Anthropic raises Haiku 5.5 to $0.50 input and $2.50 output above 100,000 input tokens. OpenAI keeps Luna at its base rate until 272,000 input tokens. Every figure below comes from the vendors' own pages, as read on 27 October 2026.

What are the list prices per million tokens?

At the base tier, the two models cost the same for input and output. They diverge only in the long-prompt tier, and the thresholds differ.

Prices as of 27 October 2026. Anthropic's Haiku 5.5 page gives the tiers in full. Anthropic's pricing page shows the same input and output rates for Haiku 5.5, but it carries no date. OpenAI's GPT-6 Luna page gives the base rates and the 272,000-token threshold.

Anthropic's table prices by prompt size: the higher rate applies to prompts over 100,000 tokens. OpenAI states its rule for the full request: above 272,000 input tokens, the input and cache rates double and the output rate rises by half.

What does a call cost on each side of the thresholds?

Below 100,000 input tokens, the two models cost the same per call. The gap opens above that point and widens with prompt size.

Costs below use list prices as of 27 October 2026. They exclude caching, batch discounts and any taxes.

The arithmetic for the 150,000-token row: Haiku charges $0.50 per million on 0.15 million input tokens, which is $0.075, plus $2.50 per million on 0.002 million output tokens, which is $0.005. The total is $0.080. Luna charges $0.10 per million on 0.15 million input tokens, which is $0.015, plus $0.50 per million on 0.002 million output tokens, which is $0.001. The total is $0.016.

At 150,000 input tokens, Haiku 5.5 costs five times as much per call as Luna. At 300,000 input tokens, Luna has already crossed its own tier, so the input rate doubles to $0.20 per million. Its bill is still about 40% of Haiku's. For a team sending 10,000 calls a day at the 150,000-token size, the list-price difference is about $800 a day against $160 a day.

How do cache reads and writes compare?

Both models make cache reads cheap at the base tier, at $0.01 per million tokens. The write prices and lifetimes differ, and the Luna page leaves one gap.

Anthropic's rates come from the Haiku 5.5 overview, which is the only page that lists the 1-hour write. OpenAI's rates come from its GPT-6 Luna page. Cache-heavy workloads should read both lists in full, since the tier thresholds for caching differ between the two.

What are the context and output limits?

Both models accept about one million tokens of context. The difference in listed input is small, and the listed output ceilings match.

The 922,000 maximum input and 128,000 maximum output on Luna add up to its 1,050,000-token window. Anthropic's Haiku 5.5 overview lists the 300K batch output in beta, which requires a specific beta header. Standard Haiku requests top out at 128K output tokens, the same ceiling as Luna. Sources: Anthropic models overview and OpenAI models page.

Do batch discounts change the comparison?

No. Both vendors halve the base price for batch work, so the gap stays the same. At the base tier, each model costs $0.05 per million input tokens and $0.25 per million output tokens on batch.

Anthropic applies a 50% discount to input and output on the Batch API. OpenAI prices Batch and Flex at 50% of its Standard rates. Haiku 5.5 also supports 300K output tokens on the Batch API, which is the one batch limit that differs. That figure is marked as beta on Anthropic's page, so it should be checked before a batch job depends on it.

Does a lower list price mean a lower bill?

Not by itself. A per-token price says nothing about how many tokens a task consumes, and tokenizers differ between models.

Anthropic's Haiku 5.5 page says the model uses a newer tokenizer. It states that the same text counts as roughly 30% more tokens than on Haiku 4.5. Neither vendor page gives a token count for the same text on the other model, so this table cannot show a like-for-like cost per task. A team that wants that number has to run its own prompts through both APIs and compare the usage fields.

The table also says nothing about output quality. It covers price and limits only. A cheaper token is a better deal only if the model completes the task in the same number of steps and with the same accuracy.

What's still unclear?

• Prompt-tier scope on Haiku 5.5. Anthropic's table says "prompts over 100,000 tokens" without spelling out whether the higher rate covers the whole request. The example costs above assume it does, which matches OpenAI's explicit rule for Luna.
• Launch date for GPT-6 Luna. The model page shows a knowledge cutoff of 18 May 2026 but no release date. This article does not give one.
• Cache-write lifetime on Luna. OpenAI's page lists $0.125 for cache writes and does not say whether that is a 5-minute or 1-hour write. Luna's 1-hour write price is not listed.
• Date stamps. Neither pricing page carries an as-of date. The prices here are as of 27 October 2026, the date the pages were read. A later price change would not appear here.
• Weakest point of this comparison. It uses list prices from a single day. Enterprise agreements, committed-use discounts and cloud-marketplace pricing are not covered.

Price changes are the most likely source of error. Both vendors update their pricing pages without notice, so the figures above should be checked against the vendor pages before a budget decision.