Anthropic's 2026 model lineup now has a clear split: Claude Sonnet 5 for most production workloads, and Claude Fable 5 for difficult work where judgment matters more than price. This guide explains the practical differences, pricing, and migration changes.
Claude Sonnet 5 is the sensible default for most teams. It brings agentic coding and tool use closer to the previous premium tier while keeping Sonnet-level pricing, making it a strong fit for coding assistants, document analysis, support automation, and other production workloads that need reliable reasoning at scale.
The model supports a native 1,048,576-token context window and up to 128K output tokens. It carries forward the tool and platform features available on Sonnet 4.6, except Priority Tier. Anthropic introduced it at $2 per million input tokens and $10 per million output tokens through August 31, 2026; standard pricing is $3/$15 afterward.
The migration detail most likely to affect cost is the tokenizer. Anthropic says the same text can produce roughly 1.0 to 1.35 times as many tokens as Sonnet 4.6. The introductory rate offsets much of that increase, but teams should measure representative prompts before forecasting spend at the standard rate.
Sonnet 5 also replaces manual thinking budgets with adaptive thinking. The model decides how much reasoning to use, which can improve difficult answers but makes per-request spend less predictable. Applications that mix complex reasoning with simple classification may still benefit from routing routine work to a lighter model.
Claude Fable 5 is the escalation path rather than the everyday route. Its strength is sustained work that combines planning, delegation, tool use, and verification over long sessions. That makes it suitable for large code migrations, complex research, and production decisions where a weak answer creates a much larger downstream cost.
Fable 5 leads several frontier evaluations, including an 86.6 score on FrontierSWE compared with 81.2 for Kimi K3 and 71.3 for GPT-5.6 Sol in the reported comparison. It also supports a 1M-token context window, adaptive thinking, mid-conversation system messages, code execution, and Anthropic's broader tool surface.
At $10 per million input tokens and $50 per million output tokens, the premium is substantial. Fable makes sense when reduced review time or lower failure risk is worth more than the model cost. Sonnet remains the better choice for routine coding and high-volume traffic.
Sonnet 5 is the lower-cost production model, while Fable 5 charges a premium for frontier capability. Opus 4.8 sits between them on price. Fable's cached input rate can materially reduce the cost of long-running agents that repeatedly reuse a stable system prompt or repository context.
| Model | Input | Output | Cached input | Context |
|---|---|---|---|---|
| Claude Fable 5 | $10.00 | $50.00 | $1.00 | 1,048,576 |
| Claude Sonnet 5 | $2.00 (promo) | $10.00 (promo) | — | 1,048,576 |
| Claude Opus 4.8 | $5.00 | $25.00 | — | 1,048,576 |
Source: Anthropic platform release notes. Sonnet 5 promo pricing through Aug 31, 2026. Sonnet 5 uses a new tokenizer producing ~30% more tokens for the same text.
Before migrating from Sonnet 4.6, replay a representative set of production prompts and compare quality, latency, token use, and cost per completed task. The updated tokenizer changes token counts even when the prompt text remains unchanged, so list price alone does not show the full migration cost.
Remove manual thinking budgets and non-default temperature, top_p, or top_k values before changing the model ID. Sonnet 5 rejects those settings with a 400 response. Adaptive thinking is enabled by default and cannot be replaced with a fixed reasoning budget.
Anthropic also raised Sonnet and Haiku rate limits to match Opus across its usage tiers, which are now consolidated into Start, Build, and Scale. Most organizations move automatically and do not need to change their account configuration.
Claude Code v2.1.198 and later runs subagents in the background by default, allowing the main session to continue with independent work and incorporate delegated results when they are ready. Teams can still control this behavior with the background frontmatter field.
Claude in Chrome is generally available for customers on direct Anthropic plans. It can open tabs, interact with pages, fill forms, and read console logs, closing the feedback loop between writing a web application and testing it in a real browser.
Claude Desktop is also available in beta for Ubuntu 22.04+, Debian 12+, x86_64, and arm64. Linux teams can use the integrated terminal, editor, visual review, and live preview workflow, though production users should validate the beta in their own environment before standardizing on it.
APItokendeal provides OpenAI-compatible access to Claude Sonnet 5, Opus 4.8, and other Anthropic models alongside GPT, DeepSeek, Gemini, and Kimi K3 — one API key, prepaid token packs, real-time usage dashboard.