Claude Haiku 5.5 Cost Calculator
Haiku 5.5 bills in two tiers: prompts up to 100K tokens pay one rate, anything above pays 5× more — decided per request, not per month. Set your real token counts and cache behavior, and this calculator shows what each request and each month actually costs.
Recount before you rebudget. Haiku 5.5 ships a new tokenizer that produces roughly 30% more tokens for the same text than Haiku 4.5. Any cost estimate measured on 4.5 — token counts, max_tokens limits, usage dashboards — has to be re-measured on 5.5 before it means anything.
All models compared at your settings
| Model | Tier at your prompt size | Per request | Per 1,000 requests | Monthly |
|---|
Same prompt/output/cache settings for every row. The highlighted row is your current pick in the calculator above.
Full published rate table
Per-million-token API list rates, verified 2026-10-17. Haiku 5.5 is the only one with two tiers; the rest bill flat.
Claude Haiku 5.5 · Anthropic · model ID claude-haiku-5-5
Released Oct 7, 2026. 1M token context, 128K max output, first Haiku with adjustable effort settings. Tier is decided by the input prompt length of a single request.
| Tier | Input | Output | Cache write (5-min) | Cache write (1-hour) | Cache read |
|---|---|---|---|---|---|
| Prompt ≤ 100K tokens | $0.10 | $0.50 | $0.125 | $0.20 | $0.01 |
| Prompt > 100K tokens | $0.50 | $2.50 | $0.625 | $1.00 | $0.05 |
Source: Anthropic launch pricing, as documented Oct 2026. Rate source ↗
Claude Haiku 4.5 · Anthropic
Previous generation, flat pricing. Included so migration budgets have an honest before/after.
| Tier | Input | Output | Cache write (5-min) | Cache write (1-hour) | Cache read |
|---|---|---|---|---|---|
| Flat | $1.00 | $5.00 | $1.25 | $2.00 | $0.10 |
Source: Anthropic API pricing (previous generation). Rate source ↗
Claude Sonnet 5.5 · Anthropic
The step up for complex agentic coding; flat pricing. Cache reads were halved to $0.10 at the same launch that introduced Haiku 5.5.
| Tier | Input | Output | Cache write (5-min) | Cache write (1-hour) | Cache read |
|---|---|---|---|---|---|
| Flat | $2.00 | $10.00 | $2.50 | $4.00 | $0.10 |
Source: Anthropic API pricing, Oct 2026. Rate source ↗
How to read these prices
The 100K line is per request. One request with a 120K-token prompt is billed entirely at the high tier — input and output. Your monthly volume doesn't matter; a single long request does. Anthropic says about 90% of Haiku 4.5 requests fell under 100K, which is why the average saving lands near 75%, not 90%.
Cache is where the real money is. Repeated system prompts and retrieved context billed at $0.01/M (cache reads) instead of $0.10/M is the cheapest lever on this page. Write new content at the 5-minute rate by default; pay the 1-hour rate only for prompts you genuinely reuse all day.
Batch halves input and output — nothing else. The Batch API takes 50% off input and output rates on all three models above (cache rates unchanged). You pay in latency instead, so it's for offline pipelines, never realtime.
Watch the tier cliff on migrations. Moving a workload from Haiku 4.5 to 5.5 cuts the rate 90% — until the 30% tokenizer inflation pushes a borderline prompt over 100K. Measure the largest real prompts on 5.5 before signing off the budget.