Anthropic Launches Claude Duet, a Lightweight Chat Model for Engineers

Anthropic has released Claude Duet, a smaller chat-optimized language model designed to run on a single GPU and accessible through a public HTTP API. Positioned as a coding co-pilot, it targets interactive assistance, data exploration, and low-latency inference rather than the heavier tasks of its larger Claude 3 sibling. The article highlights lower inference costs, since a roughly 7 billion parameter model fits in 16 GB of VRAM, along with reported response times under 300 milliseconds for 512-token prompts on a V100 GPU. It also notes that the API uses the same JSON schema as Claude 3, allowing existing tooling to be reused with minimal changes. The author walks through a sample setup that deploys a GitHub issue summarizer in Kubernetes using the official Python client.