What Is "Ultracode" in Claude Code? The High-Effort Mode That Runs Hundreds of Agents

Claude Code — Anthropic's AI coding assistant that lives in your terminal — recently gained a heavyweight mode called ultracode. People often mix it up with "ultrathink," but they're not the same thing. Ultracode is the big gun: it cranks Claude's reasoning to near-maximum and lets it spin up small armies of parallel sub-agents to chew through large engineering tasks.

Here's what it actually does, how to switch it on, where it shines — and, crucially, what it costs you in tokens.

Ultracode vs ultrathink — clearing up the confusion

This trips a lot of people up, so let's settle it first:

• Ultrathink is a per-turn nudge. Drop the word "ultrathink" into a single prompt and Claude thinks much harder on that one reply. It changes the depth of thinking for one turn, then it's back to normal.
• Ultracode is a session-wide mode. Switch it on and, for the rest of the session, Claude works at high effort and automatically orchestrates "dynamic workflows" — breaking big jobs into pieces and running many agents in parallel. It changes how Claude executes substantial work, not just how it thinks on a single turn.

In short: ultrathink is "really think about this one." Ultracode is "go run this whole project for me, properly."

What ultracode actually turns on

Flipping ultracode on does two things at once:

• xHigh reasoning effort. This is the second-highest effort level (sitting between "high" and "max") — the setting tuned for serious coding. More upfront planning, more self-checking, more edge cases caught before any code is written.
• Dynamic workflows. Claude writes its own orchestration scripts that fan a task out across tens to hundreds of parallel sub-agents in a single session, verifying the results before anything reaches you.

How to turn it on

• Session-wide: type /effort ultracode. From then on, Claude plans a workflow for every substantial task until you switch back — e.g. /effort high when you return to routine work.
• One task only: just include the word ultracode in a single prompt to run that one task as a workflow, without changing the session setting.
• It lasts for the current session and resets when you start a new one.
• It's only available on models that support xhigh effort — the Opus tier, such as Claude Opus 4.8. On other models, the effort menu won't even offer it.

How dynamic workflows work (the interesting bit)

This is what makes ultracode different from "just think harder":

• Claude breaks a large task into subtasks and distributes them across parallel agents.
• Agents tackle the problem from independent angles, with verification built in — results are checked before they're integrated.
• Progress saves continuously, so an interrupted job resumes instead of starting over.
• A single request can become several workflows in a row — one to understand the code, one to make the change, one to verify it.

Dynamic workflows launched as a research preview across the Claude Code CLI, Desktop app, and VS Code extension, and are available on Max, Team, and Enterprise plans (Enterprise needs admin enablement) as well as via the API. On Max and Team plans — and via the API — dynamic workflows are on by default.

The advantages

• Massive parallelism on big jobs. Codebase-wide bug hunts, security audits, and profiler-guided optimization passes finish in one session instead of dozens of manual back-and-forth rounds.
• Built-in verification. Agents check the work before results surface, so you get fewer wrong answers slipping through.
• Resumable. Long-running jobs survive interruptions and pick up where they left off.
• Hands-off. You state the goal once; Claude plans the workflow and runs it itself, instead of waiting for you to spell out every step.

Now the cost: what ultracode does to your tokens

This is the part most write-ups skip. Anthropic is refreshingly blunt about it:

"Dynamic workflows can consume substantially more tokens than a typical Claude Code session."

Why so much more? Two compounding reasons:

• xHigh effort means more thinking — and thinking is billed as output tokens, the expensive kind.
• Dynamic workflows multiply that across many agents — each sub-agent generates its own tokens, and one request can spin up several workflows and hundreds of agents.

To put a number on it: on Claude Opus 4.8, output tokens run about $25 per million (versus roughly $5 per million for input). A thorough ultracode session — say, a whole-codebase audit — can burn through more of a plan's weekly usage limit than a full day of normal Claude Code work.

When to use it (and when not to)

• Use ultracode for: large, parallelizable jobs — codebase-wide refactors, bug hunts, security and performance audits, anything you'd otherwise break into dozens of manual steps.
• Skip it for: everyday edits, quick questions, and small fixes. It's genuinely overkill there, and you'll pay for the overkill. For routine work, /effort high is the smart default — reach for ultracode only when the size of the job justifies the spend.

Ultracode also has cousins worth knowing: /ultraplan plans a task in the cloud before you commit to it, and /ultrareview hunts for bugs. Same "ultra" family, different jobs.

The bottom line

Ultracode is Claude Code's heavyweight mode: maximum practical effort plus an orchestra of parallel agents that plan, execute, and verify on their own. It's a genuinely powerful way to clear large engineering tasks — it just trades tokens for scale. Match it to the big jobs, keep /effort high for the rest, and you get the power without the bill shock.

Watch coding walkthroughs and AI tool demos on the Neel's World YouTube channel.