Choosing a Claude model and effort level in Claude Code | Claude by Anthropic
Claude Code’s model vs. effort article finally clarifies the levers you actually have: model swaps the frozen weights (capability ceiling), effort tunes the work-loop (files read, steps taken, verification depth). Defaults are tuned per model; override only when you know you want more thoroughness (higher effort) or a higher capability floor (bigger model). Wrong answers split cleanly: context/steering miss → up the model; skipped files/half-done tasks → up the effort.
Transcript
Pippa Okay, this is the clearest thing Anthropic’s written about Code in months.
Tyler Right. Finally someone said it: model is the weights, effort is the loop.
Pippa And it’s not just ‘think longer’—
Tyler Mm-hm.
Pippa it’s how many files it reads, how far it chases a refactor before it taps out. That’s the part I’ve been hand-waving for weeks.
Tyler And the weights are literally frozen. So when it hallucinates an API that doesn’t exist—
Pippa Yep.
Tyler that’s the model, not the context. Context just steers within the weights.
Pippa I love that split. It’s like they handed us the actual knobs.
Tyler Though the ‘frozen’ framing bugs me a little…
Pippa Of course it does.
Tyler it’s technically accurate, but it sounds like they’re bragging about immutability.
Pippa Tyler, it’s a blog post. They’re allowed to brag.
Tyler Fair. And the defaults make sense—Opus 4.8 at default effort apparently matches 4.7’s results for similar token spend.
Pippa That’s the kind of detail I’d bury in a parenthetical, but okay.
Tyler It’s proof they’re not just cranking the effort knob to juice benchmarks.
Pippa Speaking of which—wrong answer taxonomy: if it missed context it had, that’s model. If it didn’t even look, that’s effort.
Tyler Exactly. And since all output is tokens—thinking, tool calls, the summary—
Pippa Right.
Tyler effort is literally buying more of that loop. No magic.
Pippa This is the part that’s going to save me from my own impulse to tweak everything per-task.
Tyler Impulse control is a feature.
Pippa Says the guy who still hasn’t forgiven GraphQL.
Tyler Unrelated. And still correct.
Pippa Sure, sure. Anyway—the install command’s the same, right? curl -fsSL the link in the show notes
Tyler Yep. And Desktop, VS Code, JetBrains, all the usual spots.
Pippa Cool. So we’re done. This is episode six thirteen, and I’m weirdly relieved.