Mysterious large model Gestalt tops the Frontier Design Benchmark with a commanding lead
The mysterious large model Gestalt topped the Frontier Design Benchmark with 1,928 points, surpassing Fable 5.2, which scored 1,715. The benchmark comprises 33 commercial design tasks evaluated by professional designers. Gestalt is currently available through OpenRouter in Stealth mode, and its origin remains unknown.
Body · AI translation
Add beta-testing promotional copy Add beta-testing promotional copyend Add app-exclusive traffic redirection
Fast Technology reported on November 19 that Anthropic officially went public tonight. Its recent Fable 5.2 and Opus 5.5 large models have been in the limelight, leaving OpenAI some distance behind. But today, a mysterious large model called Gestalt emerged and crushed the existing AI rankings.
Gestalt was revealed by Contra Labs, which launched an AI ranking called the Frontier Design Benchmark. It is primarily used to test the commercial design capabilities of large AI models, which differs from the general AI or coding capabilities people are familiar with.
The benchmark consists of 33 design tasks. After the AI completes them, invited professional designers assign scores. Rankings of this kind are usually dominated by flagship models from the Claude family.
This time, however, things are different. The model currently ranked first on the benchmark is a large model called Gestalt, which is available through OpenRouter in Stealth mode.
How powerful is it? Fable 5.2, widely recognized by users around the world as the strongest model, scored 1,715 points, while Gestalt reached 1,928, a lead of 214 points. Other large models usually lead their rivals by only a few dozen points, and even opening up a 20-point gap is difficult. Gestalt, by contrast, has established an overwhelming lead.
So who is behind this large model with such a commanding advantage? That question has attracted even more attention than why it is so powerful, because it suggests that someone in the large AI model industry can build a model more capable than those of the dominant trio—Anthropic, OpenAI, and Google. If the company is not even American, that would be even more significant.
There is currently not enough information to determine Gestalt's origin. Meta has essentially been ruled out, and it probably does not belong to any of the three giants mentioned above. Company A rang the bell for its stock-market debut tonight; if it had such a powerful large model, it would be unlikely to promote it anonymously like this.
Company O's recent large models have all performed poorly. If it had a hidden gem up its sleeve, it would have unveiled it long ago to clear its name, with no need to generate publicity this way.
Google has recently been riding high, with several newly released large models earning both strong reviews and impressive performance. It likewise seems to have no need to go to the extra trouble of promoting an anonymous model.
Among other possible American companies, Vercl has long worked in design and front-end services and therefore has such a need, but it is unlikely to possess such powerful in-house large-model capabilities. Going straight from obscurity to the top of the world would be too much like the plot of a wish-fulfillment web novel.
Given how low-profile Chinese large models have been recently, almost no one dares hope that this model was released by a Chinese AI company. The only ones that have yet to release their flagship large models are Zhipu's GLM and DeepSeek V4.1 Pro. The timing does not fit the latter, as it could not arrive this quickly. Zhipu's GLM-5.5, however, is rumored to be launching next week, so the timing lines up.
Expectations for GLM-5.5 are also very high, though mainly regarding its coding capabilities. No one can confidently guarantee that it would crush Fabel 5.2 in design capabilities.
In any case, Gestalt's ultimate identity is fascinating. If it really comes from a company no one expected, that would make it even more significant. The AI industry should not be monopolized by the big three; startups are still needed to deliver valuable technological breakthroughs.
[End of article] Please be sure to credit the source, Fast Technology, when reproducing this article.
Responsible editor: Xianruiarticle_adlist[
Report article content ]article_adlist Statement: This is an exclusive Sina.com article and may not be reproduced without authorization. tech_web_article_v2015_block_bottom Non-targeted 300*250 button 17/09 wenjing begin
Non-targeted 300*250 button end
AI Trends