MiniCPM5-2B scores 831 on GDPval-AA v2, outpacing rivals at its size
As seen on the 24/7 Wall St. homepage on September 7, 2026.
A 2-billion-parameter model scoring 831 against a 1,000 human baseline undercuts the assumption that useful work AI requires giant clusters, and that assumption is what funds the data center buildout.
On GDPval-AA v2, which tests models on real-world work tasks against a human baseline of 1,000, MiniCPM5-2B reaches an Elo of 831, ~110 points ahead of Ling 3.0 Tiny (718) and ~180 ahead of Granite 4.2 8B (647). Models at this scale usually sit far lower, with LFM2.5-2.6B at 204
- Replies1
- Reposts0
- Likes0
Continue ReadingShow less
GDPval-AA v2 is a benchmark that evaluates AI models on real-world work tasks, scoring them against a human baseline set at 1,000. It uses an Elo-style rating system, the same kind used in chess rankings, so every point reflects head-to-head performance rather than a fixed test score.
MiniCPM5-2B, a model with just 2 billion parameters, posted an Elo of 831 on that benchmark, clearing Ling 3.0 Tiny and Granite 4.2 8B despite Granite carrying four times as many parameters.
Sponsored
_________________________________
What's Your Number...?
Here's a question most people 5y from retirement can't answer: at your current savings rate, how much do you need, and how long will it actually last? A good advisor can put a date on that in a single meeting. SmartAsset's free quiz matches you with up to three fiduciary advisors serving your area, so you can get YOUR retirement number now (sponsor)
__________________________________________
LFM2.5-2.6B, a model of comparable size to MiniCPM5-2B, sits at just 204 on the same benchmark. That spread at roughly the same parameter count shows architecture and training choices matter far more than raw model size at this tier.
Small models that perform close to human-level on practical tasks run on modest hardware rather than the massive data-center clusters the current AI infrastructure buildout assumes are necessary. If that assumption erodes, it changes the calculus for investors watching capital spending across the semiconductor and cloud sectors.