Anthropic launches new version of scaled-down 'Haiku' model | TechCrunch
Key Takeaways
- Anthropic released Claude Haiku 4.5, the newest version of its smallest model.
- Haiku 4.5 offers performance similar to Sonnet 4 at one-third the cost and more than twice the speed.
- Benchmark results show Haiku 4.5 performing competitively against models like Sonnet 4, GPT-5, and Gemini 2.5.
- The model is expected to be highly appealing for free AI plans due to its low server load requirements.
- Executives suggest Haiku 4.5 enables new production deployment styles, such as using it as a fast sub-agent alongside more sophisticated models.
Anthropic has introduced Claude Haiku 4.5, the latest iteration of its smallest AI model, which the company claims delivers performance equivalent to Sonnet 4 while being significantly cheaper and faster. Performance metrics on benchmarks like SWE-Bench and Terminal-Bench place Haiku 4.5 competitively against established models like Sonnet 4 and Gemini 2.5, particularly in areas like tool use and visual reasoning. The model's efficiency makes it highly suitable for free AI service tiers, as it minimizes server strain, and its lightweight nature facilitates parallel deployment of multiple agents. Anthropic CPO Mike Krieger highlighted that this release enables novel production architectures where Haiku sub-agents handle execution speed while larger models manage complex planning. Industry figures, like Zencoder CEO Andrew Filev, view Haiku 4.5 as unlocking new use cases, especially in latency-sensitive fields like software development. This launch follows recent high-profile releases from Anthropic, including Sonnet 4.5 and Opus 4.1.
