Anthropic Introduces Claude Haiku 4.5: Small AI Model Achieves Flagship Performance and Economy
Abstract
Anthropic released Claude Haiku 4.5 on October 15, 2025 (Wednesday ET), a smaller, faster, and more economical AI model. This model offers programming performance comparable to Claude Sonnet 4, its flagship model released five months prior, but at one-third the cost and over twice the speed. Anthropic is making the model available for free to all Claude.ai platform users, a move seen as a significant step in the competitive AI market.
Anthropic launched Claude Haiku 4.5 on October 15, 2025 (Wednesday ET), the latest compact AI model in its Claude 4 series. This model achieves the programming capabilities of Claude Sonnet 4, the top-tier model released five months ago, while costing only one-third as much and running more than twice as fast.
Claude Haiku 4.5 even surpasses Claude Sonnet 4 in certain tasks, particularly in computer operations. The model scored 73.3% on the SWE-bench Verified benchmark and 41.0% on the Terminal-Bench test, placing it on par with models like Sonnet 4, GPT-5, and Gemini 2.5.
Pricing and Availability
The API pricing for Claude Haiku 4.5 is $1 per million input tokens and $5 per million output tokens. In comparison, its predecessor, Claude Haiku 3.5, launched in October 2024, was priced at $0.80 per million input tokens and $4 per million output tokens.
Anthropic is offering Claude Haiku 4.5 for free to all Claude.ai platform users. The model is now available to developers via the Claude API, Amazon Bedrock, and Google Cloud Vertex AI platforms. GitHub Copilot has also begun rolling out Claude Haiku 4.5 incrementally to Pro, Pro+, Business, and Enterprise users.
Technical Features
Claude Haiku 4.5 is the first Haiku model to support inference capabilities, featuring a 200,000-token context window and a maximum output of 64,000 tokens (a significant increase from Haiku 3.5's 8,192 tokens), with a knowledge cutoff date of February 2025.
Anthropic suggests a multi-model collaboration approach: Claude Sonnet 4.5 is responsible for breaking down complex problems into multi-step plans, then orchestrating multiple Claude Haiku 4.5 sub-agents to complete sub-tasks in parallel. This architecture is particularly suitable for enterprise users who need to handle long-term projects.
Safety and Alignment
In Anthropic's automated alignment evaluations, Claude Haiku 4.5 demonstrated a significantly lower misalignment behavior rate than Claude Sonnet 4.5 and Claude Opus 4.1, making it the company's safest model to date. The model is classified under the ASL-2 (AI Safety Level 2) standard.
Market Competition
This release comes just two weeks after Claude Sonnet 4.5 and two months after Claude Opus 4.1. Anthropic, with a valuation of $183 billion, serves over 300,000 enterprise customers and has an annual revenue run rate approaching $7 billion this month. The company is in fierce competition with rivals such as OpenAI, valued at $500 billion, and Google.
Claude Sonnet 4.5 remains Anthropic's frontier model and the world's best coding model. Claude Haiku 4.5 offers users a new option when near-frontier performance is required, with a greater emphasis on cost-effectiveness.