Anthropic has announced an upgraded Claude 3.5 Sonnet model, a new lightweight model called Claude 3.5 Haiku, and the public beta release of "computer use" capabilities for developers.
- The upgraded Claude 3.5 Sonnet improves SWE-bench Verified scores from 33.4% to 49.0% and TAU-bench scores in retail and airline domains, while maintaining the same price and speed as its predecessor.
- Claude 3.5 Haiku matches or surpasses Claude 3 Opus on many intelligence benchmarks at a similar speed to the previous generation of Haiku, with pricing set at $0.80 per million input tokens and $4 per million output tokens.
- Computer use allows Claude to interact with computer interfaces by viewing screens and executing commands like clicking and typing; it scored 14.9% on OSWorld screenshot-only tasks, outperforming the next-best system's 7.8%.
- The new models are available via the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI, with computer use currently in experimental public beta for developer feedback.