Claude Code Field Guide
16 / 35

Models & Effort Levels

“Claude Code” refers to a family of models with different speed, intelligence and cost — and you can switch mid-conversation.

Needs correction
Currency check: The three-tier idea (fast/all-rounder/most-capable) still holds, but the specific model names and the newer “effort level” control are different from what the video describes.

The video's framing is correct at the concept level: a fast, cheap tier for simple tasks (Haiku), a reliable all-rounder for most work (Sonnet), and a top-tier model for hard problems (Opus, described in the video as “Opus 4.6”). Switch anytime with /model.

Correction: model versions move fast, and by the time you're reading this the specific version numbers in the video are dated — always check /model for what's current rather than trusting a specific version name. There's also a new top-of-line option beyond Opus in some setups, and model aliases (opus, sonnet, haiku, best) now auto-resolve to the newest available version so you don't have to track exact names yourself.

New since the video: effort levels. Separately from which model you pick, you can now tune how much reasoning it applies with /effort (low, medium, high, xhigh, max). This matters more day-to-day than most people realize — a mid-tier model at high effort can outperform a top-tier model at low effort on some tasks, and effort level is a direct lever on cost.

Matching effort to task size

A user leaves effort at the default for most work but explicitly drops to /effort low for a bulk task like renaming fifty files consistently — getting the speed and cost benefit without needing a different, cheaper model.

Reserving the expensive model for architecture decisions

A small team defaults every session to the mid-tier model, and only explicitly switches to the most capable model for a specific, hard architectural decision — keeping day-to-day spend predictable.