🛠️Claude Haiku 5.5: $0.10 Input, But Read the Tokenizer Fine Print
TL;DR
Simon Willison tests Claude Haiku 5.5 at $0.10 input and $0.50 output per million tokens. The catch: a new tokenizer uses about 1.25x the tokens of Haiku 4.5, and prices jump 5x past 100k tokens.
Simon Willison tests Claude Haiku 5.5 at $0.10 input and $0.50 output per million tokens. The catch: a new tokenizer uses about 1.25x the tokens of Haiku 4.5, and prices jump 5x past 100k tokens.
Key Points
Pricing is $0.10/$0.50 per million tokens up to 100k, then $0.50/$2.50; Haiku 4.5 was $1/$5
Same long prompt costs about 1.25x the tokens on the new tokenizer, eating part of the saving
Reasoning can't be turned off and defaults to medium; a max-effort SVG run took 5 minutes 9 seconds and cost 3.38 cents
Anthropic's own numbers show OSWorld 2.1 offline at 72.4% versus 15.7% for Haiku 4.5
Sonnet 5.5 cache reads were halved to $0.10 per million, and Max and Team plans get monthly API credits
Why It Matters
Headline per-token prices hide tokenizer changes and long-prompt tiers. Benchmark your real prompts before migrating subagent workloads.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.