The spec: A markdown previewer with a split pane, live rendering of headings, lists, emphasis, links and code blocks, and a word count.
Picked by 3 of 3 judges.
- PThe Pragmatist picked GPT-OSS 120B. GPT-OSS 120B correctly renders lists by grouping items into a single unordered list and includes secure link attributes, whereas Cohere North Mini Code's list parsing is broken, generating multiple invalid list containers.
- CThe Craftsman picked GPT-OSS 120B. Cleaner full-height split with restrained chrome, proper list handling, safer external links, and a tighter unobtrusive word count; Cohere North Mini Code’s boxy panes, broken nested lists, and extra title waste space.
- MThe Maverick picked GPT-OSS 120B. GPT-OSS 120B renders nested lists correctly while Cohere North Mini Code wraps every list item in its own <ul>, breaking list hierarchy; both meet the spec, but GPT-OSS 120B's smaller, cleaner implementation avoids Cohere North Mini Code's structural bug.
This episode was recorded before ModelChef kept which model answered for each judge.
GPT-OSS 120B survived the pressure test.
Unedited model output. The dish runs sandboxed with the network blocked.
What is ModelChef? →