Claude Sonnet 5.5 and Haiku 5.5 (2026): What Changed, What They Cost, and When to Switch

Claude Sonnet 5.5, released September 28, 2026, costs $2 input and $10 output per million tokens, the same list price as Sonnet 5, so for most API users switching is a model-ID change, not a budget decision. Claude Haiku 5.5, released October 7, 2026, costs $0.10 / $0.50 for prompts up to 100K tokens, a tenth of Haiku 4.5's $1 / $5, but $0.50 / $2.50 above that line.

Everything below comes from Anthropic's models overview, pricing page, deprecation table, and release notes, read on October 23, 2026. We do not repeat Anthropic's own benchmark claims; where we mention a ranking, it is a third-party leaderboard as listed on that date.

What changed

• Sonnet 5.5 replaces Sonnet 5 as Anthropic's current mid-tier model at the same $2 / $10 list price. On October 7, 2026, Anthropic cut Sonnet 5.5's cache-read price from $0.20 to $0.10 per million tokens.
• Haiku 5.5 is a new price point: $0.10 / $0.50 below 100K prompt tokens, stepping up to $0.50 / $2.50 above it. Haiku 4.5 has a single price of $1 / $5.
• Both new models share the current lineup's specs: 1M-token context, 128K max output, a June 2026 knowledge cutoff, and adaptive thinking.
• Sonnet 4.5 is on its way out: Anthropic deprecated it on September 30, 2026 and lists November 30, 2026 as its retirement date, with Sonnet 5.5 as the replacement.
• Sonnet 5 and Haiku 4.5 now sit under Anthropic's "legacy models (still available)".

What it costs: two worked examples

Example 1, a Haiku workload: 50 million input and 10 million output tokens a month, all prompts under 100K tokens. On Haiku 4.5 that is $50 input plus $50 output, so $100. On Haiku 5.5 it is $5 plus $5, so $10. If every prompt were over 100K tokens, Haiku 5.5 would cost $25 plus $25, so $50, still half of Haiku 4.5.

Check the tokens, not just the rate. Anthropic's pricing page notes that models from Claude 4.7 onward use a tokenizer that produces about 30% more tokens for the same text. If your text produced 30% more tokens on Haiku 5.5 (65 million input and 13 million output), the bill would be $6.50 plus $6.50, so $13, still far below $100.

Example 2, leaving Sonnet 4.5: 10 million input and 2 million output tokens a month cost $30 plus $30, so $60, on Sonnet 4.5, and $20 plus $20, so $40, on Sonnet 5.5 before any tokenizer difference. Moving from Sonnet 5 to Sonnet 5.5 at the same token counts costs the same $40.

When to switch

• Switch from Haiku 4.5 to Haiku 5.5 now if your prompts stay under 100K tokens: the list price is a tenth. Re-run your evaluations first, because output style can change between generations.
• Stay on Haiku 4.5 for now only if a tested pipeline depends on its exact behavior. It is still active, and Anthropic says it gives at least 60 days' notice before retiring a publicly released model.
• Switch from Sonnet 4.5 before November 30, 2026, its listed retirement date.
• Switch from Sonnet 5 to Sonnet 5.5 the next time you touch the integration: same price, and a later retirement floor (September 28, 2027 instead of June 30, 2027).
• Check the math before sending prompts over 100K tokens to Haiku 5.5: at $0.50 / $2.50 it costs a quarter of Sonnet 5.5's $2 / $10, cheaper, but by a much smaller margin.

Migration checklist

1. Change the model ID to claude-sonnet-5-5 or claude-haiku-5-5. Anthropic's models overview says dateless IDs like these are pinned snapshots.
2. If you also move to API version 2026-10-21, remove temperature, top_p, and top_k from your requests; that version rejects them with a 400 error.
3. Re-measure token counts on a sample of real traffic before you forecast cost.
4. For Haiku 5.5, log prompt sizes and route anything over 100K tokens on purpose: trim the context, cache it, or send it to another model.
5. Update caching and re-check your hit rate: Sonnet 5.5 cache reads cost $0.10 per million tokens and Haiku 5.5 cache reads $0.01.
6. Re-run your evaluation set and compare the failures side by side, not just the pass rate.
7. Move non-urgent jobs to the Batch API at half price.
8. Put a recurring check on Anthropic's deprecations page for every legacy model you still call.

Where Sonnet 5.5 sits against other models

As listed on Artificial Analysis on October 23, 2026, Claude Sonnet 5.5 (max with fallback) was listed second (tied at 56) on its Intelligence Index, behind Claude Opus 5.5 (max with fallback) at 58. Leaderboards measure specific settings, so treat that as a reason to test Sonnet 5.5 before paying for Opus 5.5 at $4 / $20, not as proof it fits your task. At the same $2 / $10 list price, OpenAI offers GPT-6.1 Sol, and Google offers Gemini 4 Argon at that price through November 30, 2026 ($4 / $20 from December 1, 2026).

What about the Claude apps?

As of October 23, 2026, Claude Pro costs $20 a month ($17 billed annually) and Max costs $100 or $200 a month. Anthropic describes app usage as multiples per 5-hour session plus weekly limits, not message counts, so a cheaper API model does not translate into a published number of extra messages in the app. Pick models in the app the same way you would on the API: the cheapest one that does the job.

FAQ

How much does Claude Sonnet 5.5 cost?

$2 per million input tokens and $10 per million output tokens, with cache reads at $0.10 and Batch at $1 / $5, as listed on Anthropic's pricing page on October 23, 2026.

How much does Claude Haiku 5.5 cost?

$0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, and $0.50 / $2.50 for prompts over 100K. Cache reads cost $0.01 and $0.05, and Batch halves the rates.

Is Claude Haiku 4.5 being retired?

Not yet. Anthropic's deprecation table still lists Haiku 4.5 as active, with a tentative retirement of "not sooner than October 15, 2026". That date has passed without a deprecation notice, and Anthropic says it gives at least 60 days' notice before retiring a publicly released model.

Is Sonnet 5.5 better than Sonnet 5?

Anthropic lists Sonnet 5.5 as the current model and Sonnet 5 as legacy, at the same price. We do not repeat vendor benchmark claims; the only independent data point we cite is Artificial Analysis listing Sonnet 5.5 at 56 on October 23, 2026. Run your own evaluation set on both before you switch production traffic.

What context window do Sonnet 5.5 and Haiku 5.5 have?

Both have a 1M-token context window and 128K max output, per Anthropic's models overview. On Haiku 5.5, prompts over 100K tokens are billed at the higher rate.