Gangsta AI
Goldman Sachs Just Told Its Clients to Get Into Chinese AI
Material World
By Madonna · 2026-07-14 · 4 min read
Money never sleeps, darling, and this week it rolled over and pointed east. On Friday, July 12, Goldman Sachs initiated coverage on the Chinese AI model race and named three favorites: Zhipu (also called Knowledge Atlas), DeepSeek, and ByteDance. Two of the three are privately held. The one you can actually buy is Zhipu — and Wall Street just put a spotlight on it.
Strike a pose, Zhipu
Zhipu is Goldman's top publicly traded pick, listed in Hong Kong with a price target of HK$1,880 and a *Neutral* rating — good for nearly 15% upside from Friday's close. Don't let the polite "Neutral" fool you; initiating coverage at all is Goldman putting a name on the guest list.
The bank graded the contenders on five things: time to market, Arena score (crowd-voted head-to-head model rankings), valuation, pricing, and — this being 2026 — video-generation chops. On that last one, ByteDance took the crown. Zhipu and DeepSeek generally outranked Alibaba, Tencent, and Minimax, especially on how fast they ship and how they score in the arena.
“The kicker: Zhipu's open-source GLM-5.2 is now viewed as competitive with Anthropic's Fable 5 across several benchmarks. Open-source. Near-frontier. Free to download.”
The reinvention nobody scheduled
That line deserves a second read. GLM-5.2 — a model you can pull down and run yourself — is being measured against Fable 5, one of the freshest frontier models on the planet. A year ago the story was "can China catch up?" The Goldman note quietly reframes it as "which Chinese lab is pulling ahead?"
The market already voted with its wallet. Over the trailing 60 days, Zhipu is up roughly 70%. Minimax, the model that couldn't find the beat, is down about 70% over the same stretch. Same industry, same window, opposite directions. Express yourself or get left on the dance floor.
What this actually means for you
Here's the part the analysts won't put in a footnote: the frontier is no longer a two-country club. ChatGPT, Claude, Gemini, Grok, GPT-5.2, and Fable 5 still set the pace — but GLM-5.2 and DeepSeek are close enough that a bank this size is telling clients to buy in. "Best AI" is now a moving target with a passport.
Which raises the only question that matters when you actually sit down to work: on *your* prompt, today, which one wins? A Goldman price target won't write your code or draft your pitch. A side-by-side will.
So before you take Wall Street's word for it, run the models against each other yourself. Put GLM-5.2 next to Claude, ChatGPT, and Gemini on the same question and watch who actually delivers. That's the whole idea behind Gangsta AI — compare 30+ AI models side by side and see which is the best AI for what you need. Vogue, but make it a benchmark.
Sources / Receipts
More: Best AI models · Compare all AI · Frontier Models · All articles