Claude Haiku 4.5 is Anthropic’s fastest and most efficient model, delivering near-frontier intelligence at a fraction of the cost and latency of larger Claude models. Matching Claude Sonnet 4’s performance across reasoning, coding, and computer-use tasks, Haiku 4.5 brings frontier-level capability to real-time and high-volume applications. It introduces extended thinking to the Haiku line; enabling controllable reasoning depth, summarized or interleaved thought output, and tool-assisted workflows with full support for coding, bash, web search, and computer-use tools.
Evaluations
30
across 22 benchmarks
Latency
468ms
Context length
200k
tokens
Cost
$1 · $5
input · output per 1M tokens
Key takeaways
Claude Haiku 4.5 is optimized for real-time and high-volume applications, excelling in tasks requiring swift and efficient processing, reasoning, coding, and computer-use at a fraction of the cost and latency of larger Claude models. It is a proprietary, Transformer-based model.
The model demonstrates strong proficiency in generating correct and functional Python code, with a 94.67% pass rate. It excels in direct implementation of algorithms, string manipulation, and list processing.
A significant limitation is its struggle with complex reasoning, multi-step calculations, spatial analysis, and image requirements, reflected by a very low accuracy of 4.71% on 'Humanity's Last Exam'. The model also exhibits poor calibration.
The model's poor performance on 'Humanity's Last Exam' across Biology/Medicine and Math domains is primarily due to incorrect application of formulas, misunderstanding nuanced terminology, and difficulty with questions requiring external knowledge or image interpretation.
Minor Python coding failures occurred in problems with complex or unusual mathematical definitions and subtle sorting requirements open to multiple interpretations, often struggling with recursive or inter-dependent mathematical sequences.
