Claude Haiku 4.5
Anthropic's efficiency model. Context window: 200K tokens.
Scores by test
Methodology →What you need to know
Claude Haiku 4.5 is built for high-reliability agentic workflows and complex reasoning. It achieves perfect internal scores in strategic analysis, agentic planning, and tool calling, making it an effective engine for autonomous task execution. Its 200K context window is fully utilized, as indicated by a top-tier score in long context processing.
The model provides high-end performance at a mid-tier price point. With a blended cost of $4.00 per million tokens, it delivers capabilities typically reserved for larger models—specifically in faithfulness and multilingual support—without the associated cost of flagship frontier models.
Performance is inconsistent in restrictive tasks. While it excels at planning, it struggles with constrained rewriting and exhibits a significant weakness in safety calibration. Developers should expect less precision when enforcing strict formatting constraints or requiring rigorous safety guardrails.
Use this model for agentic tool use, complex strategic planning, and processing large datasets where faithfulness is critical. Skip this model if your application requires strict adherence to constrained rewriting rules or high-sensitivity safety filtering.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models