Models, Effort and Plans

Two tasks land on your desk: renaming a prop across three files, and a race condition in payment code someone else wrote. Before reading about models and settings, decide which one deserves more of Claude's thinking.

The four models

As of November 2026, Claude Code uses Claude Opus 5.5 by default on every plan. Three other current models are a /model away. All four have a 1M-token context window and think on every step; you can't turn thinking off on them. Prices are the Anthropic API's list prices per million tokens, which is what a Console account or an Agent SDK app pays. On a subscription you don't pay per token; you work within your plan's usage limits.

Fable models are the most capable in Claude Code and suit work bigger than one sitting. The docs give four tips for them: describe the outcome rather than the steps, hand them ambiguous problems such as root-causing an outage, skip reminders to test (Fable verifies its own work more often), and give them work you'd normally split into pieces. Haiku 5.5's price rises to $0.50 / $2.50 for prompts over 100K tokens.

Switch with /model inside a session or claude --model sonnet at launch. /model saves your choice as the default for new sessions; press s on a row in the picker to switch for this session only. Two aliases are worth knowing: opusplan uses Opus while you plan and Sonnet while it executes, and default returns to your account's default.

Effort: the thinking dial

Effort controls how much the model reasons on each step. The levels are low, medium, high, xhigh and max, and all four current models default to medium, which suits everyday feature work with a clear scope. Raise it with /effort high or /effort xhigh when getting it right depends on easy-to-miss cases: a bug in code you didn't write, a flaky test, a schema change where Claude has to find every reader. Lower it with /effort low when you're iterating fast and reviewing every result anyway. Save max for a hard problem you want solved without you; the docs warn it can overthink ordinary tasks, and it lasts for the current session only.

When only one message needs deeper thought, put ultrathink in that prompt; it raises reasoning for that turn without changing your setting. Phrases like “think hard” or “be thorough” are ordinary text that nudges Claude within its effort level and doesn't raise it. In /effort or the /model picker, Enter saves a level as your default and s applies it to this session only.

Fast mode

Fast mode runs the same model with an API configuration that favors speed: up to 2.5 times faster, at a higher price per token. Quality doesn't change. It works on Opus 5.5 and Fable 5.2 but not on Sonnet or Haiku. As of November 2026 it costs $8 / $40 per million tokens on Opus 5.5 and $16 / $80 on Fable 5.2, and on subscription plans it bills to usage credits rather than your included usage. Toggle it with /fast or Option+O (Alt+O on Windows and Linux); a ↯ appears next to the prompt while it's on. Turn it on at the start of a session rather than midway, which keeps more of the conversation cached.

Plans and usage limits

Subscription usage is metered in windows: a rolling session window and a weekly one, shared across all models and with Claude chat. When you hit one, the message tells you when it resets, and switching models doesn't help. /usage shows where your usage went, broken down by skill, subagent, plugin and MCP server. Usage credits, set up with /usage-credits, let you keep working past a limit. A Console account has no windows; it pays per token at the prices above.