Z.ai Disrupts Anthropic: The AI Model Revolution in April 2026

The WebDev Arena ranking for April 2026 continues to reflect the rapid developments in the field of AI models for web development. This month, the major news is the arrival of Z.ai with its model GLM-5.1, launched on April 7. This model has directly climbed to the 3rd place in the overall ranking, disrupting Anthropic's dominance in the top three positions. Z.ai, formerly known as Zhipu AI and originating from Tsinghua University, has successfully placed an open source model under the MIT license on the podium, a first in the history of the WebDev Arena.

Anthropic narrowly misses a full sweep, but Z.ai makes its way onto the podium

Anthropic maintains a significant grip on the AI-assisted development sector. Five of the top six positions in the overall WebDev Arena ranking are occupied by Claude models: Opus 4.6 Thinking takes the lead, followed by Opus 4.6. Sonnet 4.6 ranks fourth, with the two variations of Opus 4.5 in fifth and sixth places. The only intruder in this leading group is GLM-5.1 from Z.ai, which slots in at 3rd place.

This is the first time an open source model has reached the podium of the WebDev Arena. GLM-5.1, launched by Z.ai (formerly Zhipu AI, a spin-off from Tsinghua University), is an evolution of GLM-5, which was already in 8th place in March. Another movement at the top of the ranking: Claude Opus 4.6 Thinking surpasses its standard version, reversing the order observed last month.

On the OpenAI side, the generational renewal continues. GPT-5.4 High, in its " codex-harness " variant, enters at 7th place, replacing GPT-5.2 High, which drops out of the top 10. Google keeps Gemini 3.1 Pro Preview at 8th position but loses its two models Gemini 3 Pro and Gemini 3 Flash, which exit the top 10. Alibaba makes an appearance with Qwen3.6-plus-preview, landing directly at 9th (preliminary score), while Z.ai doubles its presence with GLM-4.7 in 10th position.

HTML and React rankings confirm Z.ai's breakthrough

In HTML, Anthropic retains the top three spots (Opus 4.6, Opus 4.6 Thinking, Sonnet 4.6), but GLM-5.1 positions itself immediately behind at 4th place. Google stands out with Gemini 3.1 Pro Preview in 6th position, its best result across all categories. OpenAI sees two variants of GPT-5.4 enter the top 10: GPT-5.4 Medium in 7th place and GPT-5.4 High in 10th.

In React, GLM-5.1 confirms its breakthrough with a 3rd place, nestled between Claude Opus 4.6 and Claude Sonnet 4.6. Alibaba bursts into the top 10 with Qwen3.6-plus-preview at 7th place, ahead of GPT-5.4 High (8th). Z.ai places three models in the top 12 positions (GLM-5.1, GLM-4.7, and GLM-5), confirming its rise in the AI-assisted web development ecosystem.

Ranking criteria of the WebDev Arena

The WebDev Arena operates on the principle of blind voting. Faced with the same prompt, two models each generate a response. Users compare the results without knowing which model is behind each proposal and then vote for the one they consider the most successful. The results are then converted into an Elo score, a ranking system derived from chess and used in esports: a victory against a higher-ranked opponent earns more points, while a loss against a lower-ranked model results in a greater point loss. The ranking evolves continuously, in line with the votes.