Claude Models Compared: Which Should You Use?

Anthropic now ships five current Claude models, and picking the wrong one wastes either money or capability. Fable 5 is the top public tier for the hardest knowledge work. Mythos 5 is the same model without dual-use safeguards, restricted to approved organizations. Sonnet 5 is the mid-tier flagship for multi-agent, enterprise-scale work. Opus 4.8 is the value coder that powers Claude Code. Haiku 4.5 is the fastest and cheapest.

Four of the five are generally available today: Fable 5, Sonnet 5, Opus 4.8, and Haiku 4.5 (Anthropic). Mythos 5 is the exception — it's gated to approved organizations only (Anthropic).

This guide compares all five on price, access, and best-fit work, so you can route each task to the right model instead of defaulting to the most expensive one.

Claude Models at a Glance

The five current Claude models trade capability and access against cost. Match the model to the task, not the other way around.

All five models except Haiku 4.5 offer a 1-million-token context window; Haiku 4.5 is capped at 200K tokens (Anthropic). See our best Claude model guide for how that context-window gap plays out on real, long-document tasks.

Claude Pricing: Fable 5 vs Mythos 5 vs Sonnet 5 vs Opus 4.8 vs Haiku 4.5

Claude pricing spans a 10x range between the cheapest and most expensive current model (Anthropic). Fable 5 and Mythos 5 share the top price point at $10 input / $50 output per million tokens. Opus 4.8 costs $5 / $25. Sonnet 5 costs $2 / $10 through August 2026, rising to $3 / $15 on September 1, 2026. Haiku 4.5 is the cheapest at $1 / $5 (Anthropic).

Output tokens cost far more than input tokens on every tier, so output-heavy work — long drafting, code generation — is where the tier you pick matters most for your bill.

• Fable 5 and Mythos 5 are the most expensive current Claude models, tied at $10/$50 per million tokens (Anthropic).
• Haiku 4.5 is roughly ten times cheaper than Fable 5 per token (Anthropic).
• Sonnet 5's intro pricing ends September 1, 2026, when it rises 50% (Anthropic).
• Worked example, 5M output tokens in a month (output-only, at today's rates): Fable 5 and Mythos 5 cost about $250, Opus 4.8 about $125, Sonnet 5 about $50 (rising to about $75 after September 1, 2026), and Haiku 4.5 about $25 (Anthropic).

Access: who can use each model today

Four of the five current Claude models are generally available to any organization: Fable 5, Sonnet 5, Opus 4.8, and Haiku 4.5 (Anthropic).

Mythos 5 is the exception. It's the same underlying model as Fable 5, offered without the dual-use safety measures, to a limited set of approved organizations only (Anthropic). If you don't already know whether your organization is approved, it likely isn't — use Fable 5 instead for the same capability without the gate.

Eligibility runs through Anthropic's Project Glasswing program: vetted partners today, with a planned expansion to biomedical researchers (Anthropic). See our Claude Mythos 5 explained guide for who qualifies and how the program works.

• Generally available: Fable 5, Sonnet 5, Opus 4.8, Haiku 4.5 (Anthropic).
• Approved organizations only: Mythos 5 (Anthropic).

Which Claude Model Should You Use?

Route each task to the cheapest model that clears your quality bar, not the most capable one by default.

Use Fable 5 for the hardest knowledge work: complex software engineering, deep research, and vision tasks where a wrong answer is expensive (Anthropic). Use Opus 4.8 for general coding — it powers Claude Code and costs half of Fable 5 (Anthropic). Use Sonnet 5 for multi-agent, production-scale workflows where you're running many parallel agents (Anthropic). Use Haiku 4.5 for high-volume, latency-sensitive tasks like classification, summarizing, and routine automation (Anthropic).

Mythos 5 only enters the decision if your organization is already approved and your work specifically involves dual-use-sensitive research (Anthropic). For everyone else, Fable 5 covers the same ground.

• Hardest knowledge work, vision, research: Fable 5.
• General coding, Claude Code: Opus 4.8.
• Multi-agent, enterprise-scale production work: Sonnet 5.
• High-volume, low-latency, routine tasks: Haiku 4.5.
• Dual-use-sensitive research at an approved organization: Mythos 5.

How to use Claude

You do not host Claude yourself — you use it through a tool, so "getting started" really means choosing the right one.

The fastest way to put Claude to work day to day is inside an AI IDE, and Cursor is the most popular — it supports it directly, so you can be working in minutes. The maker's own option is Claude Code for Claude, if you want the native experience. Prefer a different editor? Windsurf, Zed, and GitHub Copilot drive these models too.