As of April 2026, Anthropic offers three model tiers designed for different workloads. Understanding their capabilities and trade-offs is essential for cost-effective production systems.
| Model | Best For | Context | Speed | Cost |
|---|---|---|---|---|
| Claude Fable 5 | Complex reasoning, frontier agentic tasks | 1M (native) | Measured | Highest |
| Claude Opus 4.8 | Complex reasoning, coding, analysis | 200K (1M beta) | Slowest | High |
| Claude Sonnet 4.6 | Balanced agentic tasks, production | 200K (1M beta) | Medium | Mid-tier |
| Claude Haiku 4.5 | High volume, low latency, classification | 200K | Fastest | Lowest |
Released May 28, 2026, Opus 4.8 introduces Adaptive Thinking — the model dynamically decides when deeper reasoning is required based on task complexity. It achieves 70% on CursorBench and 98.5% visual acuity. Substantially improved vision capabilities support higher image resolution for more accurate analysis of charts, dense documents, and complex UI screens. Note: Opus 4.8 uses an updated tokenizer that may produce 1.0–1.35x more tokens depending on content type; re-benchmark your cost estimates when migrating.
Released June 9, 2026, Claude Fable 5 is Anthropic's most capable generally available model. It represents the first "Mythos-class" model — a new tier above Opus designed for the most demanding autonomous tasks. Fable 5 features always-on adaptive thinking, a native 1M token context window, and 128K max output tokens. It is priced at $10/$50 per MTok. Fable 5 includes strict safety classifiers; queries that trigger guardrails are automatically routed to Opus 4.8 as a fallback.
Both Opus and Sonnet now support a 1 million token context window in beta. This allows analysis of entire codebases, multi-hundred-page legal documents, or massive datasets in a single request — without chunking or retrieval strategies.