Anthropic releases Claude Haiku 5.5, cuts base API price by 90%
In brief
- Haiku 5.5 API pricing: $0.10 per million input and $0.50 per million output tokens, up to 100,000 tokens
- Average saving versus Haiku 4.5 is about 75%, Anthropic estimates
- OpenAI's rival small model, GPT-6 Luna, has matching pricing, Decrypt reported
- Haiku 5.5 is the first Haiku model with an adjustable effort setting
- Claude, AWS, Google Cloud and Microsoft Azure offer it as claude-haiku-5-5
A 90% cut on paper, about 75% in practice
Haiku 4.5 charged $1 per million input tokens and $5 per million output tokens, Decrypt reported. Prompts above 100,000 tokens get a smaller 50% cut. Anthropic puts the average saving at about 75%, since roughly 90% of requests to the old model fell under the 100,000-token line (and Haiku 5.5 splits text into slightly more tokens).
So the sticker price isn't the average price.
Decrypt also reported that the new rate matches what OpenAI set for GPT-6 Luna, its rival small model. Separately, Anthropic halved Sonnet 5.5's cache-read price to $0.10 per million tokens. It's also rolling out monthly API credits this week: $100 for Max 5x subscribers, $200 for Max 20x and up to $500 pooled across users on Team plans.
What Haiku is for
Anthropic is aiming Haiku 5.5 at high-volume chores like summarizing documents and querying databases, plus speed-sensitive jobs like live customer support and operating a web browser for the user. It's the first Haiku model with an adjustable effort setting, which lets users trade cost for smarter answers.
The benchmark numbers are Anthropic's published figures, as reported by Decrypt. On OSWorld 2.1 (a test of whether an AI can operate a real computer through long, multi-step tasks, scored as partial credit), Haiku 5.5 posted 72.4% against 48.9% for Luna. On Terminal-Bench 4.0, which scores the share of professional command-line tasks an agent finishes correctly on the first attempt, it landed at 39.2%, versus 16.4% for Luna and 0% for Haiku 4.5. GDPval-AA v2.1 rates models on professional work across 44 occupations on an Elo scale; there, Haiku 5.5 scored 1620, Luna 1437 and Haiku 4.5 just 735.
Decrypt's own quick trial was less flattering. The model answered a simple logic question almost instantly, and got it wrong. That's one question from one outlet, not a benchmark.
Closing out the 5.5 lineup
Haiku 5.5 comes 15 days after Opus 5.5 (September 22) and nine days after Sonnet 5.5 (September 28), and it's the last of the three Claude 5.5 models Anthropic had promised. It's available now on the Claude website, Amazon Web Services, Google Cloud and Microsoft Azure under the name claude-haiku-5-5.
Frequently asked questions
How much does Claude Haiku 5.5 cost through the API?
Developers pay $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, according to Decrypt. That's 90% below Haiku 4.5's $1 and $5 rates. Prompts above 100,000 tokens get a 50% price cut instead.
Why does Anthropic say the average saving is about 75% rather than 90%?
Anthropic estimates the average saving at about 75% because roughly 90% of requests to the old model fell under the 100,000-token line, where the full 90% cut applies. Haiku 5.5 also splits text into slightly more tokens. Longer prompts get only a 50% cut.
Where is Claude Haiku 5.5 available?
Haiku 5.5 is available on the Claude website, Amazon Web Services, Google Cloud and Microsoft Azure under the model name claude-haiku-5-5, Decrypt reported. It's the last of the three Claude 5.5 models Anthropic had promised, after Opus 5.5 and Sonnet 5.5.
