Anthropic Dominates AI Code, Alibaba Surprises in June 2026

⚡

Key Takeaways

1Anthropic maintains its leadership position with Claude Opus 4.7, dominating the WebDevArena rankings of June 2026.

2Alibaba makes a notable comeback with Qwen3.7-max, reaching 4th place and becoming the highest-ranked non-Anthropic model.

3OpenAI drops out of the top 10, while Google makes a timid reappearance with Gemini 3.5 Flash in 10th place.

💡Why it matters — The dynamics of AI models for web development show a rapidly evolving landscape, influencing companies' technological choices.

WebDevArena: An Ever-Evolving Ranking

The WebDevArena ranking from LMArena, which evaluates AIs on front-end web development tasks, continues to transform with the release of new models. In May 2026, Anthropic had already taken the lead, and the June 2026 edition confirms this supremacy. However, this month brings two notable surprises: the return of Alibaba to the top 5 with a new model, and the disappearance of OpenAI from the top 10.

Anthropic Maintains Its Dominance

In the field of AI-assisted development, Anthropic remains the undisputed leader. The Claude Opus 4.7 Thinking and Claude Opus 4.7 models, launched in mid-April, occupy the top two spots in the ranking. They are followed by Claude Opus 4.6 Thinking in third place. In total, Anthropic manages to place five of its models in the top 10, including Claude Opus 4.6 in fifth place and Claude Sonnet 4.6 in seventh. However, Claude Opus 4.5 has dropped out of the ranking, and the brand new Opus 4.8 has yet to make its debut.

Alibaba and Rising Chinese Models

The big news this month is the return of Alibaba with its Qwen3.7-max model, which takes the fourth place. This model thus becomes the highest-ranked among those not from the Anthropic stable. Its score remains preliminary, but this performance is significant for a non-U.S. player. Alibaba is not the only Chinese representative in the ranking: GLM-5.1 from Z.ai, licensed under MIT open source, holds steady in sixth place, and Kimi-k2.6 from Moonshot occupies the eighth spot, bringing the total number of Chinese models in the top 10 to three.

Other American Players in Motion

For other American companies, the results are mixed. OpenAI has completely vanished from the top 10, with GPT-5.5 High slipping down one rank. In contrast, Google makes a quiet return with Gemini 3.5 Flash, which ranks tenth with a preliminary score as well. Meta, for its part, retains its ninth place with Muse Spark.

Here is the complete ranking of the highest-performing AI models for web development in June 2026:

Claude Opus 4.7 Thinking: 1,567 (Elo score)
Claude Opus 4.7: 1,562
Claude Opus 4.6 Thinking: 1,542
Qwen3.7-max (Alibaba): 1,541
Claude Opus 4.6: 1,538
GLM-5.1 (Z.ai): 1,533
Claude Sonnet 4.6: 1,523
Kimi-k2.6 (Moonshot): 1,518
Muse Spark (Meta): 1,508
Gemini 3.5 Flash: 1,506

The Mechanism of the WebDevArena Ranking

The principle of WebDevArena, inherited from other Arena rankings, formerly LMArena and Chatbot Arena, is based on a duel system. Two models are presented with the same request, and users vote for the most convincing response, without knowing which model is behind each proposal. These votes feed into an Elo score, a rating system inspired by chess. Winning against a higher-ranked model increases the score more rapidly, while losing to a lower-ranked model costs more points. This system allows for a continuous reshaping of the ranking, influenced by the models' performances.