“Second only to Fable 5:” Alibaba talks the talk with Qwen3.8 without providing any real data

Alibaba dropped a bombshell claim on X over the weekend: Qwen3.8, a 2.4-trillion-parameter model, is supposedly "second only to Fable 5," Anthropic's marquee system. That's a massive statement. And it arrived with zero supporting evidence — no benchmark table, no pricing, no model card, just a single tweet with a grammatical stumble that mixed up "compatible" with "comparable."

The timing is what makes this interesting. Just days earlier, rival Chinese lab Moonshot launched Kimi K3, a 2.8-trillion-parameter model that topped Arena's coding leaderboard and landed coverage on the BBC and Bloomberg. Moonshot backed its claims with an actual benchmark comparison, architecture notes, published pricing, and a concrete July 27 date for open-weight release. Alibaba, by contrast, offered vague vibes and a promise of "open-weight soon."

Here's the twist: Alibaba is Moonshot's largest shareholder, having put in roughly $800 million for a 36% stake back in 2024. So Alibaba is now publicly one-upping a company it partly owns, right as that company barrels toward an IPO. Julien Simon, an AI operating partner at Fortino Capital, calls this the "checkpoint gap" — the distance between a bold claim and the artifact that lets anyone verify it. Kimi K3's gap was tight and dated. Qwen3.8's gap is wide open, which conveniently lets Alibaba avoid two bad outcomes: publishing numbers that beat Kimi K3 and embarrassing its own investment, or publishing numbers that fall short and losing the "second only to Fable 5" bragging rights.

This also marks a reversal for Qwen, which built its reputation as one of the most prolific open-weight publishers around — Runpod's platform data even showed Qwen overtaking Llama as the most-deployed open model. But its last two flagships, Qwen3.6-Max-Preview and Qwen3.7-Max, shipped closed, API-only, no local weights in sight. Now Qwen3.8 promises to go open again, though without a date, which Simon argues functions less like a commitment and more like an option Alibaba can exercise whenever it wants — all while banking the open-weight goodwill in the meantime.

One independent test did surface Sunday, from Trilogy's Leonardo Gonzalez, pitting Qwen3.8 against Kimi K3 on a single matched task. It's a start, but a single test scored by one person on a preview model Alibaba itself admits will keep shifting isn't remotely a substitute for real benchmarks. As Gonzalez put it, the missing architecture details matter a lot for a model this size. Until Alibaba publishes something checkable, Qwen3.8 remains exactly what it looks like: a headline built to ride Kimi K3's momentum, not a verified frontier model.