AIToday
Large Language ModelsAI Business & IndustryZenn AI/MLPublished: Oct 20, 2026, 13:00 JST

Claude Haiku 5.5 ships at $0.10 input, tenth of Haiku 4.5

Claude Haiku 5.5 ships at $0.10 input, tenth of Haiku 4.5

3 Key Points

  1. What happened

    Anthropic released Claude Haiku 5.5 on October 7, 2026 at $0.10 input / $0.50 output per MTok for prompts up to 100K tokens — one-tenth of Haiku 4.5 — with 1M context and five effort levels.

  2. Why it matters

    At that price, the same budget buys far more model calls, though the new tokenizer makes the same text about 30% more tokens, so the sticker price overstates the saving.

WHO IT HITSDevelopers and small teams running high-volume, narrow-scope LLM (AI text generation) workloads — summarization, compaction, sub-agent grunt work — get a much cheaper option, but anyone with existing Haiku 4.5 integrations faces a migration that can fail silently in their own code.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

Haiku 5.5 arrives almost exactly a year after Haiku 4.5 (October 15, 2025) and completes the 5.5 generation, following Opus 5.5 on September 22 and Sonnet 5.5 on September 28. Anthropic bills it as its cheapest, fastest and smartest small model, and the pricing table backs the cheap part: $0.10 input for prompts up to 100K tokens is one-twentieth of Sonnet 5.5's $2.00. The same announcement also halved Sonnet 5.5's cache-read price from $0.20 to $0.10, which Anthropic says cuts Sonnet 5.5 costs by about 20% on most agentic tasks — a saving that reaches teams who never touch Haiku.

The migration story is where the release gets sharp. Because adaptive thinking is on by default and temperature, top_p, top_k and assistant prefill now trigger 400 errors, a model-ID swap is no longer sufficient. The author's own stock-screening tool RakuScan illustrates the trap: it reads response.content[0].text, which was always a text block under Haiku 4.5 but may now be a thinking block with no .text attribute. The resulting AttributeError slips past handlers that only catch Anthropic SDK errors, and the caller has no try block around it, so a fallback notification path never runs — the failure surfaces the next morning as a missing alert.

FAQ
How much does Claude Haiku 5.5 cost compared with Haiku 4.5?
For prompts up to 100K tokens it is $0.10 input / $0.50 output per MTok, one-tenth of Haiku 4.5's $1/$5. Above 100K tokens the rate rises to $0.50/$2.50, still half of Haiku 4.5.
Does Haiku 5.5 cost a tenth as much in practice?
Not exactly. It uses the same new tokenizer as Sonnet 5.5 and Opus 5.5, so the same text becomes about 30% more tokens than under Haiku 4.5, narrowing the gap.
Why does simply changing the model ID break existing code?
Adaptive thinking is on by default, so a thinking block can arrive first in the response. Code that reads response.content[0].text then raises AttributeError, and that error is not caught by handlers that only catch SDK exceptions.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMamdani: AI not inevitable, feds abdicated on rules