Anthropic just dropped Claude Haiku 5.5 and the pricing is actually insane
75% cheaper than Haiku 4.5 on average. We're talking $0.10 per million input tokens vs $1 before. That's a 10x cut for most workloads.
Speed? Fastest model they've ever shipped. Built for high-volume tasks like customer support, classification, database queries, browser automation.
Performance gains are wild:
• OSWorld 2.1 (computer operation): 72.4% vs 15.7% on Haiku 4.5
• Terminal Bench 4.0 (agentic coding): 39.2% (still behind Sonnet 5.5's 70.6% but that's expected for a small model)
First Haiku with adjustable effort settings so you can dial cost vs reasoning power.
Bonus moves:
• Sonnet 5.5 cache reads cut in half to $0.10 per million tokens (20% cost reduction on agentic workflows)
• Max 5x users get $100/month API credits, Max 20x get $200, Team plans get up to $500 pooled
Available now on Claude Platform, AWS, GCP, Azure. Python and TypeScript SDKs getting beta computer/browser use support.
If you're running AI agents or automation at scale, this changes the economics completely. Haiku 5.5 as a subagent alongside Sonnet for complex tasks is the play.
75% cheaper than Haiku 4.5 on average. We're talking $0.10 per million input tokens vs $1 before. That's a 10x cut for most workloads.
Speed? Fastest model they've ever shipped. Built for high-volume tasks like customer support, classification, database queries, browser automation.
Performance gains are wild:
• OSWorld 2.1 (computer operation): 72.4% vs 15.7% on Haiku 4.5
• Terminal Bench 4.0 (agentic coding): 39.2% (still behind Sonnet 5.5's 70.6% but that's expected for a small model)
First Haiku with adjustable effort settings so you can dial cost vs reasoning power.
Bonus moves:
• Sonnet 5.5 cache reads cut in half to $0.10 per million tokens (20% cost reduction on agentic workflows)
• Max 5x users get $100/month API credits, Max 20x get $200, Team plans get up to $500 pooled
Available now on Claude Platform, AWS, GCP, Azure. Python and TypeScript SDKs getting beta computer/browser use support.
If you're running AI agents or automation at scale, this changes the economics completely. Haiku 5.5 as a subagent alongside Sonnet for complex tasks is the play.