Anthropic launches Haiku 5.5 at a much lower price
The New Stack Frederic Lardinois ● Covered by 5 sources
Anthropic just cut Haiku’s price hard and gave it a big upgrade. The tiny model is now cheaper, faster to use, and far better at computer tasks.
Based on reporting by The New Stack, Frederic Lardinois — 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 on Wednesday released Claude Haiku 5.5, the first update to its smallest and cheapest model in nearly a year. It is also the company’s third 5.5 launch in a month, though Fable 5.5 is still missing and sounds like the one stuck in a slower review queue.
The headline is price. Haiku 4.5 cost $1 for input tokens and $5 for output tokens per million. Haiku 5.5 drops that to $0.10 and $0.50 for requests under 100,000 tokens. For bigger requests, the rate is $0.50 and $2.50, which is the same as Haiku 4.5. Anthropic says roughly 90% of Haiku 4.5 requests were in the cheaper bucket, and it pegs the average savings at about 75% once a newer tokenizer is included.
This version also gets effort controls for the first time, with medium as the default. That gives developers a knob for how many tokens the model uses on a task, which is a pretty direct admission that not every prompt deserves the same amount of brainpower. Anthropic is still pitching Haiku as the fastest and most efficient model, but now it is also the model that can be thrown at more work without making the bill wince.
The benchmark gains are hard to ignore. On the offline subset of OSWorld 2.1, which checks computer use, Haiku 5.5 scored 72.4%, up from 15.7% for Haiku 4.5 and ahead of GPT-6 Luna’s 48.9%. On GDPval-AA v2.1, a knowledge-work test, it scored 1,620 versus 735 for Haiku 4.5 and 1,437 for GPT-6 Luna. On Terminal-Bench 4.0, it reached 39.2%, again well above Haiku 4.5’s 0%.
Anthropic is clearly aiming this model at high-volume work like summarization, classification and routing, but it’s also pointing to compaction, database queries, live customer support and browser use. It is available on the Claude Platform, AWS, Google Cloud and Microsoft Azure, and Anthropic is adding beta support for computer use and browser use to its Python and TypeScript SDKs. The company also cut Sonnet 5.5’s cache-read price in half, and says that should make most agentic tasks about 20% cheaper.
My take — AI-written commentary, not fact-checked reporting
This is the part of the model wars that actually matters: not the biggest brain, but the cheapest one that can still do useful work. Anthropic is leaning into that reality instead of selling another poster child for hype, which is refreshing. The AI industry keeps pretending cost is a footnote; here it’s the whole story.
Read more about this at: The New Stack