Anthropic has launched Claude Haiku 5.5, a new fast and low-cost model.
The previous version, Haiku 4.5, was a year old and priced at $1 per million input tokens and $5 per million output tokens. That rate was already expensive compared to OpenAI’s GPT-6 Luna, which cost ten times less. The new Haiku matches Luna’s standard pricing of $0.10 for input and $0.50 for output, but only up to 100,000 tokens. Beyond that limit, the cost jumps to $0.50 input and $2.50 output. Luna scales at 272,000 tokens, with a lower final rate of $0.20 input and $0.75 output.
A new tokenizer makes the new model slightly less efficient. Tests show prompts use around 1.25 times as many tokens with Haiku 5.5 as they did with Haiku 4.5, adding a hidden cost. If your tasks stay under 100,000 tokens, Haiku is the cheaper option and scores higher on benchmarks. Larger tasks are better served by Luna.
The Anthropic plugin for command-line use has been updated to handle new models without requiring a new release every time. I tested the system by asking it to generate an SVG of a pelican riding a bicycle:
llm install -U llm-anthropic
llm anthropic refresh
llm -m claude-haiku-5.5 "Generate an SVG of a pelican riding a bicycle" -o thinking_effort low
Pelicans
The new model does not allow users to turn off reasoning and defaults to a medium setting. I generated images with low, medium, high, xhigh, and max effort settings. The bicycle frame looked good on all settings except the lowest. The low effort version cost 0.0936 cents and took seven seconds to produce. The max effort version took five minutes and nine seconds but cost 3.3826 cents.
Here is the pelican generated with maximum effort settings.
For comparison, the image produced by Haiku 4.5 a year ago cost 0.7583 cents. The model struggled with the drawing.
API credits for subscribers
Anthropic also cut the price of cache reads for Sonnet 5.5 by half. The company added monthly API credits to its subscription plans:
Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users.
Users claim the credits by going to Settings and selecting the API organization that should benefit from the funds. The amount matches the cost of the subscription itself, making it easier for subscribers to use the API. Anthropic allows users to disable auto-reload for the API, ensuring requests stop when the balance runs out.
Monthly credits do not roll over. OpenAI lets users apply Codex subscriptions to personal API use, which is a better deal for heavy users. This new credit scheme addresses that difference.
