In Arena's published snapshot, Gemini 4 Neon (High) ranked seventh on Code Arena's WebDev Overall leaderboard with 1,764 points, five points behind Carbon (High) and one point ahead of eighth-place GPT-6 Astra (Max). Arena's original post · Leaderboard chart
What this evaluation can assess: How Arena users compare front-end web pages and web apps generated by different models.
Limits on extrapolation: Performance and cost for maintaining existing repositories, developing back ends, running long-horizon agent tasks, or using other clients.
Model version: Gemini 4 Neon (High).
Test environment or client: Code Arena's WebDev Overall leaderboard.
Reasoning tier and parameters: The leaderboard lists the tier as High. The exact endpoint, reasoning budget, tool configuration, and sampling settings are unspecified.
Arena defines WebDev as a front-end web development evaluation. In Arena's published head-to-head voting process, users see the outputs of two anonymous models and vote for the one that better fits the request. The WebDev leaderboard shows head-to-head matchup counts and win rates, both excluding ties. It also shows confidence intervals for model strength estimated by bootstrapping. Code leaderboard categories · Arena's voting process · WebDev leaderboard
All four Gemini 4 models used the High tier. Neon trailed Carbon by 5 points and led Hydrogen by 61 points and Argon by 82 points. Arena's original post · WebDev leaderboard
At collection, the WebDev Overall leaderboard showed an update date of 2026-11-18 and a total of 1,027,059 votes. Neon's model page showed 2,308 votes, a score of 1,764 ±14, and a rank range of 3rd to 10th. On the original chart, Neon's score interval overlaps those of Carbon and GPT-6 Astra. The gaps of 5 points and 1 point between point estimates are too small to establish a stable ranking order. WebDev leaderboard · Neon's leaderboard details · Leaderboard chart
| Model and tier | WebDev Overall rank | Score |
|---|---|---|
| Claude Fable 5.2 (xHigh) | 1 | 1,824 |
| Claude Opus 5.5 (Max) | 2 | 1,818 |
| GPT-6.2 Astra (Max) | 3 | 1,797 |
| Gemini 4 Carbon (High) | 6 | 1,769 |
| Gemini 4 Neon (High) | 7 | 1,764 |
| GPT-6 Astra (Max) | 8 | 1,763 |
| Gemini 4 Hydrogen (High) | 14 | 1,703 |
| Gemini 4 Argon (High) | 17 | 1,682 |
The table draws on the chart in Arena's original post and the WebDev Overall leaderboard checked on the collection date. The chart shows the top 15. The original English post gives Argon's 17th-place rank, and the live leaderboard confirms its score of 1,682. Arena's original post · Leaderboard chart · WebDev leaderboard
WebDev's score of 1,764 comes from preference votes on web development outputs. Neon's second-place finish and 1,336 points in Agent Arena, reported in the same thread, come from a separate set of session metrics. The benchmarks cover different tasks and calculate their scores differently, so the difference between the scores has no comparative meaning. The WebDev score cannot be converted to a web page pass rate or a code correctness rate. WebDev leaderboard · Original Agent Arena post
Check the original X post and its chart for the ranks and scores at publication. To check the live data, select Code → WebDev → Overall on Arena and record the leaderboard date, vote count, score, and rank range. Reproducing the snapshot would require the full task and vote records, model invocation settings, and scoring procedure, which the post and leaderboard do not provide in sufficient detail. Arena's original post · WebDev leaderboard
Gemini 4 Neon