Claude Haiku 5.5
Simon Willison’s Weblog Simon Willison ● Covered by 5 sources
Anthropic launched Claude Haiku 5.5, a cheaper fast model. It matches Luna on price under 100k tokens, but gets pricier and uses more tokens on long prompts.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic has shipped Claude Haiku 5.5, its new fast, low-cost model, and it arrives with a very specific sales pitch: cheaper than the old Haiku, competitive with OpenAI’s GPT-6 Luna on shorter jobs, and better on benchmarks. The older Haiku 4.5 was already looking tired. It came out almost a year ago, and its $1 per million input tokens and $5 per million output tokens were high even then.
The new model is priced at $0.10 per million input tokens and $0.50 per million output tokens, but only up to 100,000 tokens. Past that point, the cost jumps to $0.50 and $2.50. Luna has its own jump at 272,000 tokens, though its higher tier is still cheaper than Haiku’s. There’s another wrinkle too: Haiku 5.5 uses a less generous tokenizer, and Simon Willison says his Claude Token Counter shows the same long prompt taking about 1.25 times as many tokens as it did with Haiku 4.5. That’s a hidden price bump, even if the sticker price looks tidy.
That makes the comparison pretty simple in practice. If your workload stays under 100,000 tokens, Haiku 5.5 is priced the same as Luna and, Anthropic says, scores higher on benchmarks. Push past that ceiling and Luna starts to look better value. The pricing story is doing a lot of the work here.
Willison also tested the model by asking it to generate an SVG of a pelican riding a bicycle. He found that the new Haiku doesn’t let you disable reasoning and defaults to medium. Low effort cost 0.0936 cents and took 7 seconds; max effort took 5 minutes 9 seconds and cost 3.3826 cents. Compared with Haiku 4.5, which cost 0.7583 cents on the same kind of test and produced a bad pelican, this is a much nicer bird.
Anthropic didn’t stop there. It also halved the price of cache reads for Sonnet 5.5 and added monthly API credits to Max and Team subscriptions on the Claude Platform. Max 5x users get $100, Max 20x users get $200, and Team subscribers get up to $500 pooled across users. The credits match the subscription cost, do not roll over, and can be managed by picking the right API organization in Settings -> Billing. Anthropic even lets users disable auto-reload so API requests stop when the balance runs out, which is a small mercy for anyone trying to avoid a surprise bill.
My take — AI-written commentary, not fact-checked reporting
This is classic Anthropic: helpful, polished, and just a little too eager to make the math interesting. The credits are genuinely useful, but the token inflation and price jumps mean buyers now need a calculator as well as a model. Closed-model companies keep saying they are making AI simpler; then they hand everyone a billing spreadsheet and call it product design.
Read more about this at: Simon Willison’s Weblog