Anthropic released Claude Haiku 4.5 on Wednesday, a small AI language model that delivers performance similar to its flagship Claude Sonnet 4 from five months ago. The new model runs at one-third the cost and more than twice the speed. All Claude app, web, and API users can access it now.
Performance and Pricing Details
According to Ars Technica, Haiku 4.5 scored 73.3 percent on SWE-bench Verified, a coding performance test. Claude Sonnet 4 scored 72.7 percent on the same test. The new model also beats Sonnet 4 at certain tasks like using computers, based on Anthropic’s benchmarks.
Anthropic ships the Claude family in three sizes: Haiku (small), Sonnet (medium), and Opus (large). Larger models run slower and cost more but typically include deeper knowledge. The company used a technique called distillation to make Haiku 4.5 match the capability of older, larger models for tasks like coding.
Cost Comparison Across Models
Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens through the API. Sonnet 4.5 costs $3 per million input and $15 per million output tokens. Opus 4.1 costs $15 per million input and $75 per million output tokens. The model is also included for subscribers of Claude web and app plans.
Use Cases and Multi-Model Workflows
Anthropic says users who need AI for real-time, low-latency tasks will benefit from Haiku 4.5. These tasks include chat assistants, customer service agents, and pair programming. The model serves as a cheaper replacement for two older models, Haiku 3.5 and Sonnet 4.
Haiku 4.5 works alongside Sonnet 4.5 in multi-model workflows. Sonnet 4.5 can break down complex problems into multi-step plans. It then coordinates multiple Haiku 4.5 instances to complete subtasks in parallel. This setup spins off workers to finish tasks faster.
The model also comes close to GPT-5 performance in certain benchmarks, though Anthropic’s self-reported results may highlight the model’s strengths. Claude Sonnet 4.5, released in late September, remains Anthropic’s best coding model. Anthropic published a system card and documentation for developers who want more details on the new model.
