T3 Code vs Cloudroom: What Happened With the ADE Tier List

On October 15, an independent tier list of ADEs (agentic development environments, the apps that run coding agents) ranked Cloudroom in God tier and put T3 Code in A tier. Theo, who makes T3 Code, then contacted the author about T3 Code's low score. Within about 30 hours, V1 was deleted, a V2 put T3 Code at the top, and Cloudroom was not on V2 at all. The next day, the same author announced a new benchmark, ADEMark, and thanked Theo for sponsoring it. Theo and the author both say that Theo never asked for Cloudroom to be removed. The author says Cloudroom was dropped because it is "still early access". Below is all the information, with links, so you can form your own opinion.

The timeline

All times are UTC.

1. October 15, 21:02. @zqkrq posts an ADE and multiplexer tier list (V1). Cloudroom is in God tier, the top tier, second of 4 tools. T3 Code is in A tier, the third tier. The post gets more than 60,000 views. It is now deleted, and the link shows "This page doesn't exist".
2. October 17, 03:21. The author posts tier list V2: "my last one wasn't fully fair, so i spent all day today retesting the latest of each on real tasks." T3 Code is now the "best tool right now". Cloudroom is not on the list. Theo reposts it 92 minutes later.
3. October 18, 15:00. The author announces ADEMark, "the first real ADE benchmark": 16 ADEs, same tasks, same machine. In the reply under it, he writes: "big thanks to @theo for sponsoring this."
4. October 18, 18:17. David Ondrej, Cloudroom's founder, posts the V1 and V2 lists side by side.
5. October 18, 18:30. The author replies to David: "Theo never told me to hide anything hahaha." He says Cloudroom "is still early access", and that Cloudroom will be in ADEMark, "same tests as everyone".
6. October 18, 19:32. Theo replies.

What Theo and the author say

Theo's reply makes four points. He contacted the author to ask why T3 Code scored low. The author had tested a T3 Code version that was more than 3 months old, then retested the new version and "adjusted". The author wanted to build a real benchmark, so Theo offered to fund it. Theo says: "I have no idea what Cloudroom is. I definitely did not tell him to remove it or anything else."

The author says that Theo "told me to be 100% transparent and that it didn't matter if t3 code ended up low". He gives this reason for V2: "cloudroom is still early access and putting it next to tools that have been on the market way longer didn't feel fair". He also says that Cloudroom will be in ADEMark, with the same tests as every other tool, and that he is "happy to show the chat".

We have no proof that Theo asked for Cloudroom to be removed, and both of them say he did not.

Why this is still a problem

The problem is the order of events. The creator of one of the tools contacts a reviewer after a low score. The next version of the list moves that tool up two tiers. The tool that was in God tier, made by a competitor who has a history of arguments on X with the founder, is not re-ranked. It is removed. Then the same tool maker pays for the reviewer's next benchmark.

Everything could have an innocent reason in theory. However, together they mean that nobody can trust V2, and nobody can trust ADEMark until the method is public. A benchmark that one of the tested vendors pays for has a conflict of interest, even if the author is honest. Model labs know this. That is why serious evals publish their tasks and their raw results, and keep the old runs online.

Two questions are still open:

• Why was "early access" a reason to remove Cloudroom from V2, when it was not a reason to keep Cloudroom out of God tier in V1, 30 hours earlier?
• Why was V1 deleted, instead of kept online next to V2?

Test both yourself

The best benchmark is your own work. Both apps are open source, and you can try both for free. T3 Code has 28,336 GitHub stars and runs Claude Code, Codex, OpenCode, Cursor and more on your own subscriptions. I tested it in my T3 Code review.

Cloudroom is an open-source desktop app (Apache 2.0) for coding agents on macOS and Linux. It runs Claude Code, Codex, Pi Agent and OpenCode in their own cloud sandboxes, on your own subscriptions. The free plan gives you 100 cloud agents and 25 hours of cloud compute every 30 days, plus unlimited local agents.

I encourage you to give both a shot and see which one you prefer.

David made a video about all of this: Theo T3 is RIGGING public benchmarks.

You can create a free Cloudroom account and test it yourself.