Claude Haiku 5.5 launched at a twentieth of Sonnet 5.5's price, with the same 1M-token context. The catch is a long-prompt tier that only Haiku has: past 100K tokens the gap shrinks from 20× to 4×.
For prompts under 100K tokens, Haiku 5.5 is 20× cheaper than Sonnet 5.5 on input, output and cache writes (10× on cache hits). Above 100K tokens Haiku bills $0.50/$2.50 per 1M, so Sonnet's $2.00/$10.00 is only 4× more — a much smaller premium if Sonnet's results save you work.
| Claude Sonnet 5.5 | Claude Haiku 5.5 | |
|---|---|---|
| Provider | Anthropic | Anthropic |
| Tier | Mid-tier | Budget |
| Input / 1M tokens | $2.00 | $0.10 |
| Output / 1M tokens | $10.00 | $0.50 |
| Cache hit / 1M | $0.10 | $0.01 |
| Cache write / 1M | $2.50 | $0.125 |
| Context window | 1M | 1M |
| Notes | Released Sep 28, 2026, replacing Sonnet 5 at the identical $2/$10 per 1M price; Sonnet 5 is now a legacy model | Released Oct 7, 2026; 90% cheaper than Haiku 4.5 with a 1M context window (for prompts ≤100K tokens); >100K tokens: $0.50/$2.50, cache read $0.05, cache write $0.625 |
Claude Haiku 5.5 is 20× cheaper on input tokens ($0.10 vs $2.00 per 1M). Claude Haiku 5.5 is 20× cheaper on output tokens ($0.50 vs $10.00 per 1M).
| Workload | Claude Sonnet 5.5 | Claude Haiku 5.5 | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $80.00 | $4.00 | Claude Haiku 5.5 saves $76.00 |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $68.00 | $3.40 | Claude Haiku 5.5 saves $64.60 |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $2.66 | $0.15 | Claude Haiku 5.5 saves $2.51 |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
the task needs more capability than Haiku delivers on your evals — complex coding, multi-step agents — or your prompts are long enough that Haiku's surcharge narrows the gap to 4×.
you run high-volume, short-prompt work such as classification, routing, extraction or support chat, where a 20× saving dominates.
Anthropic prices any Haiku 5.5 request whose prompt is over 100,000 tokens at $0.50 input, $2.50 output, $0.05 cache read and $0.625 cache write per 1M, and counts cached tokens toward that threshold. Sonnet 5.5 bills its full 1M context at standard rates. For long-document work, the honest comparison is 4×, not 20×.
With a 20× gap, even an aggressive cascade pays: run Haiku first and send only failures to Sonnet. If 10% of short requests escalate, total cost is roughly 3× Haiku-only — still more than 6× cheaper than Sonnet-only. Keep prompts under 100K on the Haiku leg (trim history, retrieve less) to stay on the low tier.
Cache hits are $0.01 vs $0.10 per 1M at base tiers — a 10× gap rather than 20×, because Sonnet 5.5's cache hits are priced at 5% of input — and both get Anthropic's 50% Batch API discount — including Haiku's over-100K rates.
20× at base rates: $0.10/$0.50 vs $2.00/$10.00 per 1M tokens. For prompts over 100K tokens, about 4×.
No. Anthropic bills Sonnet 5.5's full 1M-token context at standard rates; Haiku 5.5 is the exception.
Yes, both list 1M tokens.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.