Claude Opus 5.5 (Max) leads the WebDev Arena ranking reported for November 2026, with a score of 1,810; GPT-6 Astra (Max) is second at 1,773. Blog du Modérateur published its ranking report on November 2. The captured Arena leaderboard shows the same leading models and scores, but is marked as modified on October 27, 2026.
The ten models in the reported ranking
The November report’s top ten matches the order and scores on the captured Arena page:
| Rank | Model variant | Score | Provider |
| 1 | Claude Opus 5.5 (Max) | 1,810 | Anthropic |
| 2 | GPT-6 Astra (Max) | 1,773 | OpenAI |
| 3 | Claude Sonnet 5.5 (xHigh) | 1,770 | Anthropic |
| 4 | Gemini 4 Carbon (High) | 1,768 | |
| 5 | GPT-6.1 Sol (Max) | 1,751 | OpenAI |
| 6 | Claude Fable 5.1 (Max) | 1,746 | Anthropic |
| 7 | Muse Wonder Full (Max) | 1,725 | Meta |
| 8 | Claude Sonnet 5.5 (High) | 1,711 | Anthropic |
| 9 | Kimi K3.1 Pro (Max) | 1,695 | Moonshot AI |
| 10 | Claude Opus 5 (Max) | 1,691 | Anthropic |
The captured Arena page lists 152 models and 952,039 votes. Those figures describe that page state, which is dated October 27.
Full-stack and front-end rankings have different leaders
In the November report, OpenAI’s GPT-6 Astra and GPT-6.1 Sol lead the full-stack category. For front-end, the reported leaders are Anthropic’s Claude Opus 5.5 and Claude Sonnet 5.5, followed by Gemini 4 Carbon. The report notes that these specialized rankings received fewer votes than the overall list.
That makes the category useful when a developer’s work is mostly front-end or full-stack, but the placement still reflects this particular evaluation rather than a guarantee about every project.
What WebDev Arena evaluates
WebDev Arena ranks human preferences between anonymous website outputs. Epoch AI describes it as part of Text Arena (Coding) and documents a constrained task: produce a standalone interactive React component in TypeScript, styled with Tailwind. Its methodology description says some unstructured outputs are converted by GPT-4.1-mini before running in a Linux sandbox; applications with errors can still be shown to voters.
The score labels differ across descriptions. The November report calls the measure Elo, while Epoch AI describes a Bradley–Terry model. The captured Arena page labels its number simply “Score.” The ranking is therefore best understood as a result of preferences among outputs for a defined website-building task, not as a universal measure of coding ability or code correctness.
The walkthrough shows the earlier voting workflow
An Arena walkthrough published in 2025 demonstrates an earlier Code Arena process: two models generate a site side by side, users inspect rendered previews, vote for a preferred result, and can iterate with follow-up prompts and export code. It illustrates the comparison workflow, not the models in the reported November ranking.
