Anthropic Claude
Listed at https://console.anthropic.com
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Sonnet 5 (intro): $2/M, Opus 5: $5/M, Sonnet 5/4.5: $3/M, Opus 4.8: $5/M, Haiku 4.5: $0.80/M | per million tokens |
| Output | Sonnet 5 (intro): $10/M, Opus 5: $25/M, Sonnet 5/4.5: $15/M, Opus 4.8: $25/M, Haiku 4.5: $4/M | per million tokens |
🤖 Supported Models (7)
✨ Pros
- ✓Claude Opus 5 launched Jul 24, 2026: Opus 4.8 price ($5/$25) with near-Fable 5 performance
- ✓Claude Sonnet 5 launched: limited-time $2/M input pricing until Aug 31, 2026
- ✓Top-tier coding capabilities, preferred for Agent applications
- ✓Claude Opus 4.8 long context (200K), strongest in the Opus family
- ✓Sonnet 5 narrows gap to Opus 4.8 on agentic search/computer use benchmarks
- ✓Artifacts feature supports code previews and interactive output
⚠️ Cons
- ×API pricing still on the high end (Opus 4.8: $5 in / $25 out)
- ×China requires proxy (US export controls)
- ×Limited-time pricing window is short (Sonnet 5 only 8 weeks)
- ×Strict rate limits, especially on the free tier
🎯 Best For
Code generation and debugging; Agent/Copilot applications; Sonnet 5 limited-time pricing fits long-context + high-token workloads
💰 Pricing & Plans
| Model | Input ($/M tokens) | Output ($/M tokens) | Notes |
|---|---|---|---|
| Claude Sonnet 5 (intro) | $2.00 | $10.00 | Limited-time rate through Aug 31, 2026 |
| Claude Sonnet 5 / 4.5 | $3.00 | $15.00 | Standard rate after intro period |
| Claude Opus 5 | $5.00 | $25.00 | Top-tier reasoning, Opus 4.8 pricing |
| Claude Opus 4.8 | $5.00 | $25.00 | Long context (200K) support |
| Claude Haiku 4.5 | $0.80 | $4.00 | Fast, lightweight tasks |
🔧 API & Developer Experience
- •API Style: Distinct from OpenAI-style /v1/chat/completions — uses /v1/messages endpoint with system/user/assistant roles and a max_tokens parameter (required on every call).
- •SDK Availability: Official SDKs for Python and TypeScript/Node; community SDKs for Go, Java, Ruby. Anthropic's SDKs are the reference for building Claude integrations.
- •Tool Use: First-class tool use (function calling) with structured JSON input/output — Claude reliably handles multi-tool, multi-turn tool selection.
- •Prompt Caching: Native prompt caching reduces repeated-context cost by ~90% (cache writes priced at 1.25× input; cache reads at 0.1× input). Ideal for long-system-prompt workloads.
- •Batch API: Asynchronous Batch API at 50% discount — submit jobs that complete within 24h; ideal for bulk evaluation, dataset scoring, or non-urgent processing.
- •Streaming: SSE-based token streaming included by default; no separate Realtime API equivalent (use third-party for live voice).
- •Artifacts: Claude.ai-side feature for rendering code, HTML, SVGs in interactive preview panes — not API-exposed, but a UX differentiator for Claude.ai web users.
🧠 Coding & Agent Capabilities
Claude consistently ranks at or near the top of public coding benchmarks (SWE-bench, HumanEval+, Aider polyglot). Sonnet 4.5/5 in particular has become the de facto default model for agent-driven coding workflows — Cursor, Cline, Aider, Continue.dev, and most production coding agents default to Sonnet. The model supports multi-file reasoning, tool use for shell/file edits, and long-context sessions that fit an entire small-to-medium codebase inside the context window. Computer use (a feature that lets Claude control a browser/UI via screenshots and clicks) was pioneered by Anthropic; the latest Sonnet 5 / Opus 5 generations ship improved agentic search, file system navigation, and planning. For enterprise agent deployments, Anthropic also publishes the Claude Agent SDK (formerly Claude Code SDK) — a Python/Node toolkit for building multi-agent pipelines with native tool, memory, and sub-agent primitives.
🌐 Regional Availability & Latency
Anthropic's API endpoints (api.anthropic.com) are not directly reachable from mainland China due to U.S. export controls. Access typically requires a proxy, an overseas server, or a third-party aggregator that re-sells Claude access (such as FreeModel for Claude Sonnet/Haiku). Direct latency from China is high (500ms-2s+); aggregators often route through regional points-of-presence to bring this down to 200-500ms. For production workloads serving China-based users, an aggregator (FreeModel, OpenRouter with caveats) is the standard approach. Some developers also use Claude.ai (web interface) with a proxy/VPN for non-API use cases. Note that Claude Opus 5 is not yet re-sold by every aggregator — verify availability before migrating agents that depend on the latest model.