Key takeaways
- Anthropic's Claude Haiku 4.5 is up to 5x faster than Sonnet 4.5, enhancing coding performance.
- Achieve similar coding capabilities at one-third the cost compared to Claude Sonnet 4.
- Score 73.3% on SWE-bench Verified benchmark, outperforming Gemini 2.5 Pro and Sonnet 4.
- Haiku 4.5 is optimized for real-time tasks like chat assistance and customer service.
- Provides a cost-efficient alternative to Anthropic's flagship model, Sonnet 4.5.
Anthropic has released the latest iteration of its small AI model, the Claude Haiku 4.5.
Available for all users today, the company said that it offers similar levels of coding performance to the Claude Sonnet 4, at just one-third the cost.
On the SWE-bench Verified benchmark, which evaluates AI models on real-world software engineering tasks, Claude Haiku 4.5 scored 73.3%.
Its performance is slightly better than Sonnet 4 (72.7%) and outperforms Google’s Gemini 2.5 Pro (67.2%).
The company also shared additional benchmarks that evaluate Claude Haiku 4.5 and the competition on autonomous tasks, math, and graduate-level science and mathematics questions.
“Users who rely on AI for real-time, low-latency tasks like chat assistants, customer service agents, or pair programming will appreciate Haiku 4.5’s combination of high intelligence and remarkable speed,” said Anthropic.
While the company recently launched its flagship model, the Sonnet 4.5, it states that the Haiku 4.5 offers users a new option for “near-frontier performance with much greater cost-efficiency.”
Anthropic states that the Sonnet 4.5 model can break complex programs into multi-step tasks, leveraging several Claude Haiku 4.5 instances working concurrently to handle subtasks.
The Haiku 4.5 model’s API costs $1 and $5 per million input and output tokens, respectively. Meanwhile, the Sonnet 4.5 costs $3 and $6 per million input and output tokens, respectively.
On the other hand, Google’s Gemini 2.5 Flash, which is focused on speed and cost efficiency, is priced at $0.3 and $2.5 per million input and output tokens, respectively.
“Claude Haiku 4.5 delivers intelligence without sacrificing speed,” said Ben Lafferty, a staff engineer at Shopify, in a testimonial to Anthropic.
In another testimonial, Andrew Filev, CEO of Zencoder, said Haiku 4.5 runs four to five times faster than the Sonnet 4.5 model.
Co-founded and led by former OpenAI executive Dario Amodei, Anthropic’s Claude family of models, along with products like Claude Code, has gained significant popularity within the software development community for its coding capabilities.
