What it is
Anthropic sells API access to its Claude models. NanoClaw runs Claude through Anthropic's Claude Agent SDK by default and Nanobot has a native Anthropic provider; OpenClaw, Hermes Agent and ZeroClaw can call it too.
Strengths
- Native provider in NanoClaw (through the Claude Agent SDK) and in Nanobot; the other agents call it directly or through OpenRouter
- Prompt caching cuts the price of a repeated prefix, such as the system prompt and the tool list an agent resends at every step, by up to 90% per Anthropic's documentation
Limits
- Usually dearer per token than small open models served elsewhere
- Prompts leave your hardware, as with any hosted model
Best for
- Agent loops that call many tools
- Long documents and codebases
Not for
- Cost as the first criterion
- Data that must stay on your hardware (use a local model)
Our take
Our opinion, not a benchmark: Claude is a sound default for agent loops that call many tools. Turn on prompt caching for the system prompt and the tool list, which an agent resends on every step, and set a spending limit in the console before you leave an agent running unattended. Prices change with each model generation, so read them on the pricing page rather than from a figure copied here.
Sources
Read on 2026-10-27.
Alternatives
- OpenAI (GPT API): OpenAI's API for the GPT models. Its chat format is the one that OpenAI-compatible servers and gateways copy.
- OpenRouter: One OpenAI-compatible API in front of many model providers, with fallback between them.