model comparison

vs

Claude Haiku 5.5 vs Claude Sonnet 5.5

two Anthropic models, compared on listed price, real workload cost, context and capabilities.

the short answer

  • on Terminal-Bench 4.0, as each maker reports it, Claude Haiku 5.5 scores 39.2% and Claude Sonnet 5.5 70.6%.
  • Claude Haiku 5.5 lists lower input and output prices.
  • Claude Haiku 5.5 reads cached input for less ($0.01/M against Claude Sonnet 5.5's $0.1/M), which decides the bill for agents that resend long context.

side-by-side specs

specClaude Haiku 5.5Claude Sonnet 5.5
providerAnthropicAnthropic
input price$0.1/M$2/M
output price$0.5/M$10/M
cached input$0.01/M$0.1/M
blended price (3:1)$0.2/M$4/M
context window1M tokens1M tokens
max output128k tokens128k tokens
inputtext, imagestext, images
structured outputyesyes
weightsclosedclosed
releasedOctober 7, 2026September 28, 2026
Terminal-Bench 4.0 (maker-reported)39.2%70.6%
Humanity's Last Exam (with tools) (maker-reported)57.4%64.5%

same work, both models

workloadClaude Haiku 5.5Claude Sonnet 5.5
one chat reply3,000 tokens in, 500 outClaude Sonnet 5.5 costs 20× as much as Claude Haiku 5.5$0.00055$0.01
summarise a 100-page report50,000 tokens in, 1,500 out, one request$0.0058$0.12
1,000 classification calls500 tokens in, 20 out, per call$0.06$1.20
coding-agent session2M tokens in across ~60 turns, 60,000 out, no cache$0.23$4.60
same session, cached85% of input read from the prompt cache, the rest written to itClaude Sonnet 5.5 costs 18× as much as Claude Haiku 5.5$0.08$1.52

at listed rates. token counts are typical sizes, not measurements.

Claude Haiku 5.5 pricing notes

  • every rate rises 5x once a prompt passes 100K tokens, counting cache reads and writes; the listed price is the sub-100K rate.
  • cache writes bill at $0.125 and cache hits at $0.01 per 1M tokens below 100K prompt tokens.
  • its tokenizer counts the same text as about 30% more tokens than Claude Haiku 4.5.

Claude Sonnet 5.5 pricing notes

  • the full 1M-token window bills at the standard rate, with no long-context surcharge.
  • cache writes bill at $2.50 per 1M tokens (5-minute tier); cache hits at $0.10, half Sonnet 5's $0.20.

when to pick which

Claude Haiku 5.5

Anthropic's small Claude model, with adaptive thinking, image input and a 1M-token context. it is the low-cost tier below Sonnet 5.5, priced at a tenth of Haiku 4.5's rates for prompts up to 100K tokens.

pick it if

  • you run high-volume classification, extraction, routing or subagent tasks
  • your prompts stay under 100K tokens, where every rate is lowest
  • you resend long prompts, so cache hits bill at a tenth of input

skip it if

  • you need complex agentic coding; Anthropic points to Sonnet 5.5 or Opus 5.5
  • your prompts routinely pass 100K tokens, where every rate is 5x

Claude Sonnet 5.5

Anthropic's mid-tier Claude model, which Anthropic describes as its best combination of speed and intelligence. it succeeds Sonnet 5 at the same $2 / $10 rate, with a 1M-token context and half the cache-read price, and sits below Opus 5.5.

pick it if

  • you want Claude at half of Opus 5.5's per-token price
  • you resend long prompts; cache hits bill at 5% of input
  • your prompts need a 1M-token window at the standard rate

skip it if

  • Haiku 5.5 handles your high-volume classification or routing more cheaply
  • your code forces tool use or sets temperature; both return errors

faq

which is better, Claude Haiku 5.5 or Claude Sonnet 5.5?

on Terminal-Bench 4.0, as each maker reports it, Claude Haiku 5.5 scores 39.2% and Claude Sonnet 5.5 70.6%. these are each maker's own numbers, not a controlled head-to-head; the spec table lists every benchmark both publish.

which is cheaper, Claude Haiku 5.5 or Claude Sonnet 5.5?

Claude Haiku 5.5 lists lower input and output prices. on a coding-agent session (2M tokens in, 60,000 out) Claude Haiku 5.5 costs $0.23 and Claude Sonnet 5.5 $4.60.

can i switch between Claude Haiku 5.5 and Claude Sonnet 5.5 mid-task?

yes. in an octomind session, /model octohub:claude-haiku-5-5 and /model octohub:claude-sonnet-5-5 switch models without losing the conversation, so you can draft on one and review on the other.

try either model

switch between models inside the same octomind session.

prices from the catalog of 2026-11-18