Claude Haiku 4.5 goes legacy: we drop it to conditional as Haiku 5.5 lists at a tenth of the price

What happened
Anthropic now lists Claude Haiku 4.5 as legacy and names Claude Haiku 5.5 the current Haiku, with Haiku 4.5 still at $1 / $5 per 1M tokens.
Why it matters
Haiku 5.5 lists at $0.10 / $0.50 per 1M up to 100K with a 1M context, so we move Haiku 4.5 from recommended to conditional.
What to do
Start new work on Haiku 5.5, test it on your workload, and plan the migration of existing Haiku 4.5 integrations now.

Our verdict: Claude Haiku 4.5 drops from recommended to conditional. It still works and its price has not moved, but Anthropic now lists it as legacy and names Claude Haiku 5.5 the current Haiku. At list price the successor costs a tenth as much per token on prompts up to 100K, so Claude Haiku 4.5 is no longer the value baseline we rated it as. Keep it for existing integrations while you migrate; do not start new work on it.

What happened

Anthropic's Claude Haiku 4.5 model page now opens with "Legacy" and says that, although the model is still available, you should consider migrating to Claude Haiku 5.5. The page lists:

  • Status: Active (legacy)
  • Retirement: "Not sooner than October 15, 2026", with no firm shutdown date
  • Price: $1 input and $5 output per 1M tokens, unchanged
  • Limits: 200K context window, 64K max output
  • Thinking: manual extended thinking (budget_tokens), not adaptive thinking

Why it matters

We rated Haiku 4.5 recommended because it was Anthropic's value baseline. Anthropic's own pricing page breaks that framing:

Claude Haiku 4.5Claude Haiku 5.5
Input per 1M tokens$1$0.10 up to 100K, $0.50 over 100K
Output per 1M tokens$5$0.50 up to 100K, $2.50 over 100K
Context window200K1M
Max output64K128K
ThinkingExtendedAdaptive

Even on prompts over 100K, Haiku 5.5 lists at half the Haiku 4.5 rate, with five times the context.

Two caveats keep this from being a pure price story:

  • Tokenizer. Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, while Sonnet 4.6 and earlier use the previous one. Per-token prices therefore overstate the gap, though not by enough to close a tenfold difference.
  • Retirement floor. The "not sooner than October 15, 2026" date has already passed. Anthropic has not published a shutdown date, but nothing in its commitment now stops one being scheduled.

What changes for you

  • New projects: start on Haiku 5.5. Our directory rates it conditional: every benchmark is Anthropic's own, and prompts over 100K cost five times the base rate.
  • Existing Haiku 4.5 integrations: they keep working, but plan the migration now. Anthropic links a Haiku 4.5 to Haiku 5.5 migration guide from the model page.
  • Before you switch: recount token budgets for the newer tokenizer, and check your thinking settings, since Haiku 4.5 uses manual extended thinking and Haiku 5.5 uses adaptive thinking.
  • Test first: we have not benchmarked the two models against each other, so check quality on your own workload before cutting over.

FAQ

Does Claude Haiku 4.5 stop working now?

No. Anthropic lists it as Active (legacy) on every platform it supported. The retirement floor has passed, though, so treat continued availability as time to migrate rather than a long-term guarantee.

Is Haiku 5.5 cheaper in every case?

At list price, yes. Up to 100K tokens per prompt it is $0.10 input and $0.50 output per 1M, against $1 and $5 for Haiku 4.5. Over 100K it rises to $0.50 and $2.50, still half the Haiku 4.5 rate.

Claude Haiku 4.5 logoClaude Haiku 4.5

recommendedprevious pick
conditionalnew pick

Anthropic lists Claude Haiku 4.5 as legacy and names Claude Haiku 5.5 the current Haiku, which lists at $0.10 / $0.50 per 1M tokens up to 100K against $1 / $5.

Affected tools & models

Never need to catch up again

The weekly delta: only verdict changes and act-now items. No digest filler.

By subscribing you agree to our Privacy Policy. Unsubscribe anytime.