Only second-rate models do anonymous benchmark bragging

"Bull's Coming, Guess Your Horse"

Not hard to see why I'm writing this — mainly got burned by the "Bull's Coming" model.

Starting last Thursday night, the "Bull's Coming" LLM dropped with earth-shattering fanfare, sending hordes of secondary-market bros scrambling around asking: which company is behind the Bull's Coming model?

Come on. If you're buying model-company stocks, you must believe in AGI. How are you trading on rumors? Spending all day "listening at the bedside" — you're just getting rekt by the market makers.

It's that time of the month again: guess-the-model season. Seriously, how do you all keep letting these bloggers who love playing along with anonymous model hype toy with you? I hate riddlers the most.

Hard to believe it's late August 2026 and we're still doing anonymous leaderboard drops for people to guess who made it. The whole "Bull's Coming" model — guess your 🐎 what?

What's there to guess, fam? If they named it "Bull's Coming," they definitely want the stock market to soar. So 99% it's Zhipu AI, with the remaining 1% in case they suddenly refuse to claim it.

And you gotta admit, "Zhipu Bull's Coming" — rolls right off the tongue, auspicious as hell, like they were made for each other from birth.

And now that they can't top the leaderboards against Fable, they've started pulling moves like "exceeds Fable and GPT-5.6 on select benchmarks" — there's always some angle to claim a win, huh?

In case anyone forgot, Zhipu already pulled this anonymous model leaderboard stunt when they launched GLM-5 earlier this year.

They dropped a model called "Pony Alpha" on OpenRouter, free to use, rounded up a bunch of Chinese Twitter bloggers to guess who it was. Had everyone playing along: was it DeepSeek-V4? Grok 4.2?

Then Zhipu revealed the answer themselves, manufacturing this "holy shit, GLM-5's coding ability is near Claude Opus 4.5" moment — which Guancha.cn called "China's AI two bombs, one satellite moment."

So now, I hereby declare this "China's LLM Bull's Coming moment." Same old trick, works every time. Going further and further down the old-timer road.

Running through this year's anonymous model leaderboard enthusiasts:

• Xiaomi's "Hunter Alpha" model;
• Ant Group's "Elephant" model;
• Meituan's LongCat "Owl Alpha";
• And Alibaba's video model Happy Horse, so sneaky it fooled their own people.

(Feel free to add more)

The hype playbook's basically identical. Let me summarize the classic anonymous distribution path:

Anonymous model release, OpenRouter free-for-a-limited-time, guaranteed top-three placement — get some people on Twitter to benchmark it, gasp that certain results beat Claude and GPT — Chinese self-media repackages the Twitter content as proof the model's amazing — finally the real identity gets revealed, basking in traffic.

Hoping Zhipu claims the Bull's Coming model ASAP 🌹

Honestly want to round up all you anonymous LLM people (though Xianyu says LongCat can do whatever, support domestic models and chips), pure waste of everyone's attention.

The most disgusting part is they also plant bloggers to guess it's DeepSeek. Come on, DeepSeek's way too straight-laced to pull this kind of thing — only hype companies love doing this.

I also thought carefully about which other model shop besides DeepSeek is straight-laced, never done anonymous models.

First one that came to mind: StepFun. Not sure if StepFun will go even harder than Zhipu post-IPO, not sure if StepFun simply can't figure out anonymous models, but StepFun is indeed steadily releasing second-tier models.

Seriously, my suggestion: treat any company that loves anonymous models as a second-tier model shop. If a domestic model actually had the muscle of Qwen 3.8 Max or K3-level big-parameter performance, why stay anonymous? They'd definitely rush out a "shocking debut" before post-training is even finished. The ones taking it slow, free for a week, gradually building hype — clearly not SOTA.

Yep, anonymous leaderboard drops are the passport of second-tier models.

I posted on Jike before: next one to pull an anonymous model stunt on everyone, I'm sniping. Four months later, here we are again.

Looking forward to the next manufacturer that still loves anonymous models ❤️

Final disclaimer: not saying there's anything wrong with this "Bull's Coming model," and from rumors it seems to likely be a small model — if it actually delivers on the hype, definitely usable, free is free after all.

But if you're going to announce, announce earlier. Getting it into people's coding plans sooner beats everything.

My point is, good models like Qwen 3.8 27B that skip the leaderboard games, announce directly, and actually deliver "intelligence for the price of cabbage" are rare. This is what "frontier intelligence belongs to everyone" actually looks like.

That said, if the Bull's Coming model actually pumps Zhipu's stock, maybe consider paying some licensing fees to the Bull's Coming director?

(Cover image generated by ChatGPT, purely human-written)

⬇️

Subscribe to our Substack: funeralai.substack.com