There's also a tokenizer efficiency difference: modern Claude's 100K tokens are about ~60-65K modern GPT tokens, so in reality the Luna cutoff is much further away than the Haiku one.

You can test with Anthropic's count_tokens endpoint or with https://crates.io/crates/tokwc

▲

AtNightWeCode 6 hours ago | parent [-]

> ...this tokenizer, the same input text produces approximately 30% more tokens on Claude Haiku 5.5 than on Claude Haiku 4.5.

So, it is might be even worse.

	▲	Tiberium 5 hours ago \| parent [-]
		No, it's just Haiku 4.5 is so old that it predates the new Claude tokenizer change in Claude 4.7+