Anthropic API Cheat Sheet

Updated 2026-04-29

The Anthropic API provides programmatic access to Claude, Anthropic's family of advanced large language models, enabling developers to integrate state-of-the-art AI capabilities into applications. Available through direct API access, AWS Bedrock, Google Vertex AI, and Microsoft Foundry, the API supports text generation, vision, function calling, streaming responses, server-side tools, and agentic workflows. As of April 2026, Claude Opus 4.7 is the flagship model with a 1M token context window, adaptive thinking, and step-change gains in agentic coding; the 1M context window is now standard pricing with no surcharge across all Claude 4.x models. The API follows a stateless message-based architecture, and new offerings like Claude Managed Agents (public beta April 2026) provide full hosted infrastructure for production agent deployments at $0.08 per session-hour.

What This Cheat Sheet Covers

This topic spans 14 focused tables and 112 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Claude ModelsTable 2: Core API FeaturesTable 3: Request ParametersTable 4: Authentication & ConfigurationTable 5: Tool Use & Function CallingTable 6: Server-Side ToolsTable 7: Streaming & Real-Time OutputTable 8: Vision & Multimodal InputTable 9: Context & Memory ManagementTable 10: Cost OptimizationTable 11: Error Handling & ReliabilityTable 12: SDK & IntegrationTable 13: Advanced FeaturesTable 14: Production Best Practices

Table 1: Claude Models

Model	Example	Description
Claude Opus 4.7	`model: "claude-opus-4-7"` `1M context, 128K output`	• Current flagship GA model (released April 16, 2026) • step-change in agentic coding, vision, and knowledge work • adaptive thinking only (no manual budget_tokens) • temperature/top_p/top_k removed (400 error if set) • new tokenizer uses up to 35% more tokens than prior models • $5/$ 25 per million input/output tokens.
Claude Sonnet 4.6	`model: "claude-sonnet-4-6"` `1M context, 64K output`	• Best speed/intelligence balance; 1M context GA as of March 13, 2026 • 30–50% faster than Sonnet 4.5 • supports adaptive thinking and extended thinking • $3/$ 15 per million tokens • default for most production use cases.
Claude Haiku 4.5	`model: "claude-haiku-4-5"` `200K context, 64K output`	• Fastest and most cost-effective at $1/$ 5 per million tokens • supports extended thinking • no prompt injection protection — review before agentic use • ideal for high-volume, low-latency tasks.
Claude Opus 4.6	`model: "claude-opus-4-6"` `1M context, 128K output`	• Previous flagship; still available at same price ( $5/$ 25) • supports adaptive + manual extended thinking • fast mode beta available (6× price) • superseded by Opus 4.7 for most use cases.

Table 1: Claude Models

Model	Example	Description
Claude Opus 4.7	`model: "claude-opus-4-7"` `1M context, 128K output`	• Current flagship GA model (released April 16, 2026) • step-change in agentic coding, vision, and knowledge work • adaptive thinking only (no manual budget_tokens) • temperature/top_p/top_k removed (400 error if set) • new tokenizer uses up to 35% more tokens than prior models • $5/$ 25 per million input/output tokens.
Claude Sonnet 4.6	`model: "claude-sonnet-4-6"` `1M context, 64K output`	• Best speed/intelligence balance; 1M context GA as of March 13, 2026 • 30–50% faster than Sonnet 4.5 • supports adaptive thinking and extended thinking • $3/$ 15 per million tokens • default for most production use cases.
Claude Haiku 4.5	`model: "claude-haiku-4-5"` `200K context, 64K output`	• Fastest and most cost-effective at $1/$ 5 per million tokens • supports extended thinking • no prompt injection protection — review before agentic use • ideal for high-volume, low-latency tasks.
Claude Opus 4.6	`model: "claude-opus-4-6"` `1M context, 128K output`	• Previous flagship; still available at same price ( $5/$ 25) • supports adaptive + manual extended thinking • fast mode beta available (6× price) • superseded by Opus 4.7 for most use cases.