Anthropic Shaves Claude Haiku 5.5 Prices by Up to 90%

Anthropic's cheapest model just got cheaper

Anthropic has released Claude Haiku 5.5, the company's cheapest, fastest, and most capable small model yet, aimed squarely at high-volume, cost-sensitive work such as classification, extraction, routing, and subagent tasks. It lands nine days after Claude Sonnet 5.5 and completes the Claude 5.5 family that began with Opus 5.5 in late September.

Where it sits in the Claude 5.5 family

Haiku 5.5 is the small workhorse of the lineup: it carries much of the efficiency headroom of Sonnet 5.5 but is tuned to do many more jobs for far less money. Anthropic says it costs around 75% less to run than Haiku 4.5 on average, while a June 2026 knowledge cutoff keeps it current enough for production pipelines that need recent facts.

• A 1,000,000-token context window with up to 128,000 output tokens (300,000 on the Batch API in beta)
• $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens
• A five-fold price step above that threshold and a new tokenizer that counts roughly 30% more tokens for the same text
• Adaptive thinking on by default with an adjustable medium effort setting
• Available in the Claude API as claude-haiku-5-5 and on Amazon Bedrock

The price cut is the story

For the short prompts that made up roughly 90% of traffic to the previous Haiku model, input now costs $0.10 per million tokens and output $0.50, a per-token cut of about 90% versus Haiku 4.5's $1 and $5. That lands exactly alongside OpenAI's smallest model, GPT-6 Luna, which carries the same price up to its own step threshold.

Anthropic is not stopping at the small model. It is halving the cache-read price for Sonnet 5.5 to $0.10 per million tokens, a move it says trims typical agent workloads by roughly 20%, and rolling out monthly API credits of up to $500 for Max and Team subscribers. All of it arrives as Anthropic pushes toward an initial public offering expected before Thanksgiving.

What this means for you

For developers, Haiku 5.5 makes cheap, large-context subagents and batch pipelines dramatically more economical. If your workload is mostly short prompts, the new tier is nearly free relative to last year's cost while reading five times as much context at once. The caveat: prompts over 100,000 tokens step up in price five-fold, and the updated tokenizer counts about 30% more tokens for the same text, so re-benchmark your real jobs before assuming the top-line savings.

Sources: Anthropic: Introducing Claude Haiku 5.5; Simon Willison's Weblog; Quartz.

Corrections: Report an error at corrections@wireshaman.com.