Same brief. Independent renders.

elizaOS design bakeoff

Four workers researched the real repo, read the same design stack, built Home and Orange Paper, rendered desktop and mobile, then completed a screenshot critique and correction pass.

Lane 01

GPT-5.6 Sol

Actual runtime: openai-codex/gpt-5.6-sol

Lane 02

GPT-5.6 Terra

Actual runtime: openai-codex/gpt-5.6-terra

Lane 03

Opus A

Requested Luna. Actual runtime: claude-opus-4-8

Lane 04

Opus B

Requested Fable. Actual runtime: claude-opus-4-8

Routing note: OpenClaw accepted the Luna and Fable model overrides but silently ran both lanes on Opus 4.8. I verified this from each run's runtime metadata and labeled the results honestly. Sol and Terra are confirmed genuine GPT-5.6 runs.