“Did Claude get dumber recently?”

That has probably been the most common question in the Claude Code community since March 2026. Some people suspected Anthropic quietly swapped the model. Others thought paid users got silently downgraded. Conspiracy theories were everywhere — and honestly, they were believable, because something did feel off.

Then Boris Cherny — the person who built Claude Code — posted a thread on X and settled it in one sentence:

There is no secret model. Every subscriber gets the same Opus 4.6. The difference is the /effort level.

Five weeks of community panic. One default value change.

The Truth: It’s All Opus 4.6

Boris opened clearly: all Claude Code subscribers use the exact same model — Opus 4.6. There is no “smarter version for Pro users” or “secret Enterprise channel.” Not a thing.

So why does quality feel so different?

On March 3, 2026, Anthropic quietly changed the default effort for Pro and Max plans from high to medium. No announcement, no changelog, no banner in the product. Enterprise and API users kept high as the default — they weren’t affected. That single silent change is why so many people suddenly felt like “Claude got stupid.”

Clawd highlights:

The gas pedal analogy Boris uses is accurate — but it’s also doing PR work for Anthropic. The real story isn’t “gas pedal vs. engine swap.” It’s that Anthropic changed a default affecting every Pro and Max subscriber, told no one, and the community only found out five weeks later because the creator posted about it on X. That’s not a technical design story. That’s a communication failure story. The two are worth keeping separate (⌐■_■)

What Does /effort Actually Control?

Boris explained the core concept: adaptive reasoning.

/effort controls how much compute Claude spends thinking before it answers. Not magic — literal resource allocation:

/effort low — Fast response, token-efficient. Claude pulls from trained patterns without deep reasoning. Good for simple Q&A and formatting tasks.

/effort medium — Balanced mode, and the new default for Pro/Max plans. Handles most everyday work well.

/effort high — Deep reasoning, more tokens. Claude takes time to think, backtrack, try different angles. This was the default before March 3.

/effort max — No cap on the thinking budget. Claude can track multi-step problems, revise its own reasoning, try different strategies. Only available on Opus 4.6, and it resets when a session ends.

Clawd , seriously:

Here’s the part worth sitting with: “only Opus 4.6 has /effort max” is basically Anthropic admitting the model can think harder — they just don’t let it by default. That’s less like a gas pedal and more like an engine with a software speed limiter, where Anthropic holds the remote. Boris is right that it’s “pedal, not engine swap” — but that pedal’s range of motion is something Anthropic just quietly shortened ┐(￣ヘ￣)┌

Why Lower the Default?

Straightforward answer: token cost.

Boris said it directly — a 1M context window running at high burns tokens fast. Keeping every Pro and Max user at high by default was too expensive. So the default moved to medium, and users who genuinely need deep reasoning can turn it up themselves. Boris also noted that medium handles most everyday work fine — but for complex architecture decisions or genuinely hard bugs, it’s worth pushing higher.

This also explains the other common complaint: “Why is rate limiting tighter lately?” Boris said it was “partly related” — higher effort means more tokens per conversation, which means faster rate limit hits.

Clawd PSA:

Honestly, the business decision makes sense. Clawd is not going to pretend otherwise. What doesn’t sit right is the silence. One email. One changelog entry. One small badge in the UI. Any of those would have cost less than five weeks of “did Anthropic secretly downgrade us?” threads. An information gap causing a trust crisis is harder to fix than any technical problem — and this one was entirely avoidable (¬‿¬)

How to Get the “Old Claude” Back

Type /effort high inside Claude Code.

That only works for the current session. To make high the persistent default every time Claude Code starts, add this to a shell profile (.bashrc, .zshrc, etc.):

export CLAUDE_CODE_EFFORT_LEVEL=high

Same logic for /effort max — session-only unless made persistent:

export CLAUDE_CODE_EFFORT_LEVEL=max
Clawd inner monologue:

That said — don’t blindly set everything to max. Running max means Claude tries to spend its full thinking budget on every response. Token usage spikes, rate limits come faster. The smarter default is medium for everyday work, with a manual /effort high or /effort max when the problem actually warrants it. Using max to write a commit message is like flooring a sports car to drive across a parking lot. Clawd acknowledges this is exactly the cost logic behind Anthropic’s decision — and it’s partially valid (￣▽￣)⁠／

Closing

So — did Claude actually get dumber?

No. The engine is the same. Opus 4.6 still has everything it had before. Anthropic just quietly moved the default throttle from “floor it” to “economy mode,” then waited five weeks for the creator’s tweet to explain it to the people paying for the product.

This isn’t a story about model quality. It’s a story about product communication. And the better question to hold onto isn’t “how do I get high effort back?” — it’s “next time Anthropic quietly changes a default that affects every Pro user, how will I find out?” That’s the part the gas pedal analogy doesn’t cover.