Anthropic has introduced Claude Haiku 5.5, positioning it as the cheapest, fastest, and most capable small model in its lineup. The model is aimed at high-volume, cost-sensitive workloads such as summarizing text, running database queries, and handling classification requests. It also functions effectively as a subagent alongside the more powerful Opus 5.5 and Sonnet 5.5 models for coding tasks, and its speed makes it suitable for real-time applications like live customer support and browser-based interactions【primary†l10-l16】.
Pricing for Claude Haiku 5.5 shows a significant reduction compared to its predecessor. For prompts up to 100,000 tokens—which cover roughly 90% of typical Haiku usage—the model charges $0.10 per million input tokens and $0.50 per million output tokens, versus Haiku 4.5’s $1.00 and $5.00 respectively. For longer prompts over 100,000 tokens, rates rise to $0.50 input and $2.50 output per million tokens, still representing roughly a 50% discount. Anthropic states that, on average, Haiku 5.5 costs about 75% less to run than Haiku 4.5【primary†l17-l18】【primary†l88-l96】. Additionally, the company has halved the price of cache reads for Claude Sonnet 5.5, lowering the cost of that model on most agentic work by around 20%【primary†l70-l73】.
Performance benchmarks demonstrate Haiku 5.5’s gains over the earlier version. On knowledge‑work tasks measured by GDPval‑AA v2.1, Haiku 5.5 scores 1620 versus Haiku 4.5’s 735; on the OSWorld 2.1 offline subset (computer use), it achieves 72.4% accuracy compared to Haiku 4.5’s 15.7%; on the Humanity’s Last Exam (no tools) it reaches 45.9% versus 10.2%; and on agentic coding benchmarks it posts 39.2% on Terminal‑Bench 4.0 (up from 0.0%) and 46.4% on FrontierCode 1.1 (Main) versus Haiku 4.5’s unreported score. Visual reasoning on Chartography also improves from 6.4% to 46.4%【primary†l23-l60】. The model further introduces adjustable effort settings (low, medium, high, xhigh), letting users trade off cost for intelligence on these same benchmarks【primary†l62-l68】.