micjm models developers’ preferences on Agent Duel, where coding models compete head to head. In the fitted analysis, response speed appears about as important as model identity. That makes a faster Gemini interesting, even before fresh comparisons arrive.

  • The author models win rates using model identity, speed, early voting, and how speed interacts with early voting.
  • Assuming Neon matches Carbon’s capability at the reported 127 tokens per second, its projected rating is 1,066 versus Opus 5.5’s 1,078.
  • The projected chance of beating Opus is 48.3%, counting ties as half a win. The 90% interval for the rating difference runs from -94 to +68.

Caveat: this is a projection from Carbon data, not measured Neon results; the speed relationship also comes from observational data.

What the thread adds

The discussion on Hacker News clarifies the voting method and narrows one workflow comparison:

  • seizethecheese, identifying themselves as the author — “The hitch is there’s no real Neon data.” The same reply specifies “head to head battle data by real developers on top of OpenCode.”
  • seizethecheese explains early voting: “We let users vote for the finished agent without waiting for the agent that’s still working.” They also clarify that a “developer vote” is simply a vote, rather than a separate category. This makes willingness to wait part of the preference being measured.
  • rootusrootus narrows their Copilot–Claude Code comparison: “I haven’t done enough actual coding with GHCP yet to get a feel, so my original comment may be too broad.” Their repeated comparisons concern code review in both command-line clients. They report finding more bugs with Claude Code; this remains a personal observation.
  • theturtletalks — “The other Gemini models looked good initially and in benchmarks, but on long-running sessions and orchestration, it’s fallen short.” Their report concerns earlier models, rather than a controlled Neon comparison.

Handles are pseudonymous, HN exposes no per-comment scores, and ordering is HN’s ranking; this is a slice of the thread, not a consensus.