Claude Code Ultraplan and Ultrareview: Cloud Power Tools
Claude Code ultraplan and ultrareview explained: plan big changes in the cloud, review them in your browser, then catch bugs with a multi-agent review.
Evgenii Arsentev · PhDReviewed for accuracy by Evgenii Arsentev, PhD · 2026-10-14
Anthropic removed the Ultraplan research preview in Claude Code v2.1.222. The /ultraplan command, the ultraplan keyword trigger, and the Refine with Ultraplan plan-approval dialog option are all gone. For planning, use plan mode in your local session (/plan), or start a Claude Code on the web session directly. Ultrareview (/code-review ultra) is unaffected and works as described below.
Ultraplan and ultrareview are Claude Code's two cloud power tools: /ultraplan drafts a plan for a big change in a Claude Code on the web session while your terminal stays free, and /code-review ultra launches a fleet of reviewer agents in a cloud sandbox to find verified bugs before you merge. One plans the work, the other checks it.
I'm Evgeny, and I'll be honest: these two features sound like they're for engineering teams, but they solve a very non-programmer problem — I can't tell if the AI's plan is good, and I can't tell if its code is safe. Ultraplan gives you a plan you can read and comment on in a browser like a Google Doc. Ultrareview gives you a bug report where every finding was independently reproduced. Both are training wheels for judgment you don't have yet.
What is ultraplan in Claude Code?
Ultraplan hands a planning task from your local CLI to a Claude Code on the web session running in plan mode. Claude researches your codebase and drafts the plan in the cloud while you keep working in the terminal. This was a research preview — it has since been removed. See the callout above for alternatives.
What is ultrareview, and how is it different from a normal review?
Ultrareview is a deep code review that runs on Claude Code on the web infrastructure: a fleet of reviewer agents explores your change in parallel inside a remote sandbox, and every reported finding is independently reproduced and verified — so you get real bugs, not style nagging. Compare that to the local /review: seconds to minutes, single pass, counts toward normal usage. Ultrareview now takes roughly 15 to 25 minutes. It runs in the background while you keep working, and it's built for pre-merge confidence on substantial changes.
/code-review ultra
Reviews the diff between your current branch and the default branch, including uncommitted changes. To review a GitHub pull request instead: /code-review ultra 1234. The old name /ultrareview still works as an alias. Track or stop it with /tasks.
How much does ultrareview cost?
Pro and Max subscribers get 3 free runs as a one-time allotment that does not refresh, and after that each review bills to usage credits — typically $5 to $25 depending on the size of the change. Team and Enterprise plans get no free runs. A run counts once the cloud session starts, so a review you stop early still spends a free run. Before launching, Claude Code shows a confirmation with the scope, your remaining free runs, and the estimated cost — no surprise charges. Ultrareview also requires signing in with a Claude.ai account; an API key alone will not do.
For a quick is this okay while iterating, the local /review is faster and counts as normal usage. Save ultrareview for changes where being wrong is expensive — pre-merge on substantial features, or when you have no senior engineer to sanity-check the output.
The kicker: before you ship a change that makes you nervous, spend one free ultrareview run. That bug report, independently verified, is the closest a non-programmer gets to having a senior team on call.
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →References
Selected peer-reviewed papers and preprints on the AI research behind this topic.
- [1]Espejel et al. (2023). A Comprehensive Review of State-of-The-Art Methods for Java Code Generation from Natural Language Text. arXiv:2306.06371
- [2]Luo et al. (2023). WizardCoder: Empowering Code Large Language Models with Evol-Instruct. arXiv:2306.08568
- [3]Shende et al. (2026). Perceptual Self-Reflection in Agentic Physics Simulation Code Generation. arXiv:2602.12311
- [4]Ranaldi et al. (2024). Self-Refine Instruction-Tuning for Aligning Reasoning in Language Models. arXiv:2405.00402
- [5]Sapkota et al. (2025). Object Detection with Multimodal Large Vision-Language Models: An In-depth Review. arXiv:2508.19294
- [6]Jang et al. (2026). Self-Refining Video Sampling. arXiv:2601.18577
- [7]Zucchet et al. (2025). How do language models learn facts? Dynamics, curricula and hallucinations. arXiv:2503.21676
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Reading is a good start
Want to actually build this?
Guides explain. The free course transforms — personalized, gamified, and built to get you shipping fast.
Start the free course →