Anthropic launches Claude Haiku 5.5, cutting price to match OpenAI's GPT-6 Luna

Anthropic has released Claude Haiku 5.5, a faster, lower-cost language model that replaces Haiku 4.5, which came out almost a year ago. Developer Simon Willison wrote on his blog that the new model’s pricing appears aimed squarely at a recent OpenAI release.

Haiku 4.5 cost $1 per million input tokens (chunks of text an AI model processes) and $5 per million output tokens, a price Willison called expensive even at launch. Haiku 5.5 cuts that to $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, matching the price OpenAI set last month for its GPT-6 Luna model. Beyond that threshold, Haiku 5.5’s price rises fivefold, to $0.50 and $2.50 per million tokens. Luna also rises in price beyond 272,000 tokens, but only to $0.20 and $0.75.

Willison flagged a hidden cost increase: Haiku 5.5 uses a new tokenizer (the system that splits text into countable units) that is less efficient than the one Haiku 4.5 used. Running the same long prompt through his Claude Token Counter tool, he found it used about 1.25 times as many tokens under Haiku 5.5. A prompt that looks cheaper on paper can end up costing close to the same once that difference is counted.

Willison tested the model with his llm-anthropic plugin, using his standard benchmark prompt asking it to draw a pelican riding a bicycle as an SVG image. Haiku 5.5 does not allow reasoning (the model’s step-by-step internal processing) to be switched off, and defaults to a “medium” setting. He said the low-effort version produced a weak bicycle frame, while every setting above low drew a good one. The low-effort pelican cost 0.0936 cents and took seven seconds; the maximum-effort version took five minutes and nine seconds but cost only 3.3826 cents. For comparison, he said the pelican Haiku 4.5 produced a year ago, which did not support reasoning levels at all, cost 0.7583 cents and “sucked” at drawing pelicans.

More credits for subscribers

Alongside Haiku 5.5, Anthropic said it is halving the price of cache reads (reused portions of a prompt billed at a lower rate) for its Sonnet 5.5 model. It is also adding monthly API credits to its Claude subscription plans: Max 5x subscribers get $100 a month, Max 20x subscribers get $200, and Team subscribers get up to $500 pooled across their users. Anthropic said the credits match the cost of the subscription itself and do not roll over if unused. Subscribers can claim them from Settings, then Billing, by selecting which API organization should receive the credits.

Willison noted that OpenAI still lets Codex subscribers put their subscription toward personal API use, which he said remains a better deal for heavy API users, though Anthropic’s new credit scheme narrows that gap somewhat.

What happens next depends on whether Haiku 5.5’s heavier tokenizer quietly erodes the headline price cut for longer prompts, and whether OpenAI adjusts GPT-6 Luna’s pricing or efficiency in response.