Chinese models are very close to saturating CritPt (it's broken and in reality 30% is…
This is a AI post classified by Jev as Frontier models (an opinion), kept by the AI Radar because it carries real work, not commentary.
Chinese models are very close to saturating CritPt (it's broken and in reality 30% is more like 90%). This RL run had cost peanuts, 5 days of a ≈10K GPU cluster, it didn't even target CritPt explicitly. We need harder evals. Real world R&D hard.
Posted by Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) (76.4k followers) 1 h ago · 56 likes · 3.6k views · view the original post on X. Kept by the AI Radar as Frontier models. Tools mentioned: Intelligence Benchmarking.
More AI work like this
- Grok 4.7 just landed — and Terminal-Bench nearly doubled. — @davidarngar
- I was too quick and excited when landed in SF. As most of you pointed out the real deal… — @kimmonismus
- MiMo V2.6 pro is on par with Grok 4.7, yet a fraction of the price — @MaaSonder
- By the way. — @teortaxesTex
- 🥵 — @mervenoyann
- XIAOMI 🔥: MiMo-V2.6-Pro and Flash open-source models have been released! — @testingcatalog
- Damn. I’m late to the party. Plane just arrived in SF: Grok 4.7 looks super amazing! — @kimmonismus
- just taking a moment to appreciate how far these 9B models have come. — @suchenzang
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 40.3k posts from 5k X accounts over the last 14 days, 1.7k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-21 22:29 UTC. Full method.