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
11
across 11 benchmarks
Latency
8.6s
Context length
200k
tokens
Cost
$1 · $5
input · output per 1M tokens
Key takeaways
Claude Haiku 4.5 is a proprietary Transformer-based model by Anthropic, optimized for real-time and high-volume applications at lower cost and latency than larger Claude models.
The model is a Transformer-based architecture that supports controllable reasoning depth, summarized or interleaved thought output, and tool-assisted workflows including coding, bash, and web search tools.
Claude Haiku 4.5 excels in reasoning, coding, and computer-use tasks, demonstrating strong proficiency in generating correct Python code (93.33% Pass@1). Its limitations include struggles with interpreting subtle problem constraints, unusual decoding/sorting rules, complex sorting keys, and inferring highly optimized built-in functions without explicit guidance.
Aime 2025: Claude Haiku 4.5 (Reasoning) achieved an accuracy of 26.67% on AIME 2025, excelling in number theory and direct algebraic manipulation but struggling with geometry, combinatorics, and intricate algebraic expansions due to calculation errors or misinterpretation.
Python Coding Challenges: Claude Haiku 4.5 (Reasoning) achieved a 93.33% Pass@1 rate on Python Coding Challenges, demonstrating high proficiency for a variety of programming tasks when examples are straightforward and align with common algorithmic patterns.
