Four open-weight models over 30B ship permissive after January 2026 — GLM-5.2 leads GPQA at 91.2%

Asked: which open-weight AI models have a permissive license, more than 30B parameters, a release date after January 2026, and a reported GPQA score — with all four attributes for each?

A live-web scan found four distinct qualifying models (Tencent "Hy3" and "Hunyuan 3" normalized as one), released April–July 2026, from 31B to 744B total parameters, all reporting GPQA Diamond scores of 84.3–91.2%.

Circle area = total parameters (B); inner disc = active MoE parameters.MIT licenseApache 2.0GLM-5.2, highest reported GPQA and largest model

GLM-5.2 is both the largest (744B total, 40B active) and the highest scorer, at a reported 91.2% GPQA-Diamond under MIT — computeleap.com

Gemma 4 31B is the only dense model in the set and the smallest at 31B, yet still reports 84.3% GPQA Diamond — daily.dev

DeepSeek V4 Flash gets 88.1% GPQA Diamond from just 13B active parameters, the leanest MoE routing of the four — demandsphere.com

Tencent's Hunyuan 3 and "Hy3" are the same 295B / 21B-active Apache 2.0 model, reported at 90.0% GPQA Diamond by two separate outlets — wan27.org, techtimes.com

Model	Developer	License	Total params (B)	Active params (B)	Released	Reported GPQA	Benchmark label	Sources

Live-web scan as of 2026-08-30. Five scanned candidate rows normalized to four models (Tencent "Hy3" merged into "Hunyuan 3"). Criteria: MIT or Apache 2.0 license, more than 30B total parameters (not active MoE parameters), release date 2026-02-01 or later, and a reported GPQA score. GPQA figures are vendor- or press-reported GPQA Diamond results, not independently reproduced, and evaluation settings can differ between sources; decimal-precision scores taken from the source pages where available. An April 2026 "Hy3 Preview" entry (87.2%) was superseded by the July release figure.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.