Claude

A family of large language models built by Anthropic, with a focus on reasoning, coding, and safety. Reachable through claude.ai, the Anthropic API, and the apps and agents built on top, including Claude Code. The vault's agent integration runs on Claude.

Model tiers

The current family is 4.x, split into three tiers that trade capability against speed and cost:

• Opus: most capable, for hard reasoning and agentic work. Latest is Opus 4.8.
• Sonnet: balanced, the everyday workhorse. Latest is Sonnet 4.6.
• Haiku: fastest and cheapest. Latest is Haiku 4.5.

A 1M-token context window is available on the larger models.

Capabilities

• Tool use: structured function calling, the basis for agentic behaviour. See Tools.
• MCP: speaks the Model Context Protocol to reach external services.
• Extended thinking: a budget of reasoning tokens before answering on hard problems.
• Prompt caching: reuse a stable context prefix across calls to cut cost and latency.
• Vision, citations, batch, files: multimodal input plus production-API conveniences.

Notes

• Trained with Constitutional AI; safety and refusal behaviour are deliberate design goals, not afterthoughts.
• Without subscription subsidies, heavy agentic use gets expensive fast, the same cost wall that Agents hit in general.