Anthropic Releases Claude Haiku 5.5: A Small Model With 1M Context Priced at $0.10 per Million Input Tokens
MarkTechPost Michal Sutter ● Covered by 16 sources
Anthropic launched Claude Haiku 5.5, a tiny model with a 1M-token window and very low prices. It’s meant for volume work, not front-line coding, and Anthropic says it’s faster and far cheaper than Haiku 4.5.
Based on reporting by MarkTechPost, Michal Sutter — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic has pushed out Claude Haiku 5.5, and the pitch is simple: keep the model small, make it fast, and price it for jobs where the tokens fly. The company says it is now its cheapest and quickest Haiku model, aimed at summaries, classification, compaction, and subagent work rather than heavyweight reasoning chores.
The headline numbers are hard to miss. Haiku 5.5 keeps a 1M-token context window and can generate up to 128K output tokens. On list price, it starts at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens. Anthropic says that is 90% below Haiku 4.5 for that prompt range, and after accounting for the new tokenizer it estimates the newer model runs about 75% cheaper on average.
There’s a catch, though, and it matters if you live inside pricing tables. The rates jump above 100K prompt tokens to $0.50 per million input tokens and $2.50 per million output tokens. Anthropic also says the new tokenizer counts the same text as roughly 30% more tokens than Haiku 4.5, and non-default temperature, top_p, or top_k values will return a 400 error. Batch jobs are in beta with support for up to 300K output tokens, and batch processing cuts another 50% off.
Haiku 5.5 is generally available as a hosted API across the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. It takes text and images, outputs text, and ships with adaptive thinking on by default and the effort setting at medium. Anthropic says it has a June 2026 knowledge cutoff.
The benchmark story is better than you’d expect from a model that is supposed to be the small, cheap one. Anthropic reports 72.4% on OSWorld 2.1 offline subset, 39.2% on Terminal-Bench 4.0, and 46.4% on FrontierCode 1.1 Main. Sonnet 5.5 still leads those rows, and Anthropic itself says Sonnet 5.5 and Opus 5.5 are the better choice for complex agentic coding. So Haiku 5.5 looks less like a flagship and more like a practical worker bee, which may be exactly the point.
My take — AI-written commentary, not fact-checked reporting
This is the kind of launch that actually makes sense: a small model with a real price cut, not another vendor trying to win the benchmark Olympics with a bill attached. Anthropic is also being honest about where it fits, which is refreshing in a market that usually slaps “best” on everything. The boring model that saves money and finishes the job is often the one that matters.
Read more about this at: MarkTechPost