Anthropic split one question across 5 Claude agents and beat its own best single model by 90.2%

Same question. Same model family. The only change was how many were reading at once. ▸ PLAN - Opus 4 reads the question and writes the plan ▸ SPAWN - splits it into 3-5 subagents,

mugenSep 264

Mushroom Rally Update: Tux the Linux penguin rides along 🐧 and the weather changes every lap ⛈️❄️🛸

Everything in the browser, no download. A joint work of Blender, Unreal, ElevenLabs, Opus 5, Opus 5.5, Astra 6 & GLM 5.3. Play:

Martin BeSep 26

5 AI agents built a working product with payments and tests in 30 minutes.

The whole run cost $3.70. The trick isn't the models. It's the order. repo → data model → API → UI → evals No layer starts until the one under it is closed. Who does what: → the operator on Haiku

WoodySep 2611

I also tried this prompt 👀 ​​Claude Code (Opus 5.5 + HyperFrames),

I read the memory and Obsidian and rattled it. Gather materials, compose and publish 15 second version. The tempo was so fast that I couldn't read it, Tell Opus5.5 about any concerns you may have. This is the 34 second version with content adjustments.

みやしょー｜セールスデザイナー・AI活用パートナーSep 268