Claude Haiku 5.5
Anthropic's smallest model, and a brilliant update: smart enough for most text work and a lot of scripting, at a fraction of the price of the bigger models.
Quality
Modality
multimodal
Context
1M tokens
Access
closed
Fabian's Take
"This is a brilliant update, and I think people underestimate it. It's quite smart: anything around manipulating text, and even some scripting, it handles for a fraction of the cost of the mid-tier and top models. I don't pick it by hand. It runs as a subagent whenever the orchestrating agent is confident it's good enough, and in our automations, for example to summarize and translate customer chatbot conversations. For internal translations it's perfect; for anything customer-facing I'd reach for a smarter model."
Claude Haiku 5.5 is the smallest model in the Claude 5.5 family, released on 7 October 2026, two weeks after Opus 5.5 and nine days after Sonnet 5.5. Anthropic built it for high-volume, latency-sensitive work: classification, extraction, routing and subagent tasks.
Why it’s worth a second look
Haiku 4.5 had fallen so far behind that I only used it to search large document collections. 5.5 is a different model. It handles most work that’s about text, from summaries and translations to rewriting and pulling data out of documents, plus a fair amount of simple scripting, and it does it fast.
The price is what makes it interesting. Up to 100,000 tokens of prompt, a million input tokens cost 10 cents and a million output tokens 50 cents: a twentieth of Sonnet 5.5 and a fortieth of Opus 5.5. Above 100,000 tokens of prompt it costs five times as much, still well below Sonnet.
Where I use it
Not by picking it. In my agent sessions, the main agent hands focused jobs to subagents, and Haiku gets the ones the orchestrating model is confident it can do. And in automations, where the same small job runs over and over: at work, conversations from our customer chatbot are summarized and translated by Haiku. For internal use that’s all you need. For text a customer will read, I’d still use a stronger model.
It replaces Haiku 4.5, which Anthropic now lists as a legacy model.
The Verdict
Best for: Subagents and automations: summarizing, translating, extracting, rewriting, and simple scripting at high volume.
Pros
- A big step up from Haiku 4.5, and smarter than most people expect from the smallest model
- Handles almost anything that manipulates text, and some scripting too
- A twentieth of Sonnet 5.5's price per token for prompts up to 100K tokens
- 1M-token context and 128K output, like the rest of the Claude lineup
Cons
- Prompts over 100K tokens cost five times as much
- For customer-facing translation and writing, a stronger model is the safer choice
- It counts the same text as about 30% more tokens than Haiku 4.5 did
Specs
- Pricing $0.10/M input, $0.50/M output for prompts up to 100K tokens ($0.50/$2.50 above) · cache read $0.01/M · batch 50% off
- Cost Tier budget
- ⚡Speed Tier instant
- License Proprietary
