Haiku 4.5: fast, accurate tool calling
Anthropic's Claude Haiku 4.5 - now available in Madgwick - delivers breakthrough performance for AI assistants that use tools, with speeds up to twice as fast as larger models.1
For staff building assistants that integrate external capabilities - whether web search, data analysis, or custom functions - Haiku 4.5 represents a significant upgrade in both reliability and responsiveness.
Tool calling without the wait
The most immediate benefit of Claude Haiku 4.5 is speed. When your assistant needs to call a tool - searching the web, analysing data, or executing a function - Haiku 4.5 completes these operations in a fraction of the time of previous models. In direct comparisons, it completes tool-calling tasks 2-4 times faster than Claude Sonnet 4.5 for tool-dependent workflows.2
The model demonstrates particularly strong performance in:
Parallel tool orchestration: Calling multiple tools simultaneously without confusion or dropped requests.
Multi-step reasoning: Planning around tool results more effectively, reducing the brittle behaviour that plagued earlier models.
First-attempt accuracy: Getting tool calls right the first time, significantly reducing the need for retry loops.3
Intelligence at a lower cost
Haiku 4.5 scores 73.3% on SWE-bench Verified4 - a rigorous coding benchmark - matching the performance of much larger models. It delivers this capability at approximately one-third the cost of Claude Sonnet 4.5, meaning your Madgwick usage stretches further.5
When to choose Haiku 4.5
Claude Haiku 4.5 is well-suited as a default model for most tool-enabled assistants in Madgwick.
It excels at:
Web search assistants: Fast, reliable integration with search capabilities.
Data analysis tools: Quick execution of analytical functions without sacrificing accuracy.
Custom function assistants: Any assistant that needs to call external capabilities reliably.
For assistants that don't use tools, or for tasks requiring the deepest possible reasoning on highly complex problems, other models in Madgwick's range remain available. But for tool-calling reliability, speed, and cost-efficiency, Haiku 4.5 represents a significant advancement.