Ten open-weight models pass the screen; GLM 5.2 leads at 91.2% GPQA, and scale does not map cleanly to score

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? List all four attributes for each model.”

A corroborated web screen (as of 2026-09-07) surfaced 10 developer-released open-weight models with total parameters above 30B, release after 2026-01-31, an MIT or Apache 2.0-family license, and a numeric GPQA Diamond result. These 10 are all models surviving the screen, not a canonical global registry. Scores span 71.2% to 91.2%; parameter counts are total (including total MoE parameters, not active).

MIT licenseApache 2.0 licensex-axis is log scale · circle size follows corroborating hosts

GLM 5.2 tops the screen at 91.2% GPQA Diamond with 753B total parameters — less than half the size of DeepSeek V4 Pro (1,600B, 90.1%). local-ai-zone.github.io

Scale does not map cleanly to score: 35B Qwen 3.6-35B-A3B (86.0%) beats 119B Mistral Small 4 (71.2%) and nearly matches 754B GLM 5.1 (86.2%). computingforgeeks.com

Licenses split 4 MIT to 6 Apache 2.0; all four MIT entries are Chinese labs (Z.ai, DeepSeek) and hold three of the top four scores. qtithow.com

Corroboration is uneven: GLM 5.2 appears on 64 independent hosts while Qwen3.5-35B-A3B and DeepSeek-V4-Pro-0813 rest on just 2 hosts each. hokai.io

All 10 screened models

Model	Developer	Total params (B)	Released	License	GPQA Diamond %	Hosts	Source

Method: web research across roughly 4,000 search results — official model cards, repositories, papers, announcements and independent benchmark sources — as of 2026-09-07. 10 rows shown, one per screened model. Screen: total parameters > 30B (total, not active), initial public release after 2026-01-31, MIT or Apache 2.0-family license, numeric reported GPQA Diamond score (normalized to %). Community fine-tunes, merges, third-party repackagings, previews and use-restricted licenses excluded; models kept only with developer identification or corroboration on at least two hosts. Month-only source dates shown as the first of the month (approximate, marked ≈). Click a column header to sort.

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.