Nine open-weight models over 30B parameters shipped permissive licenses after January 2026 — GLM-5.2 leads at 91.2 GPQA-Diamond

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.”

Nine distinct qualifying models, released between 11 February and 6 July 2026, drawn from 13 evidence rows across 15 source pages as of 2 September 2026. Sizes span 31B (Gemma 4 dense) to 1.6T total parameters (DeepSeek V4 Pro); GPQA-family scores run from 68.2 to 91.2, as reported — variants and evaluation settings differ by source.

MIT family (incl. Modified MIT)Apache 2.0Mistral permissive research + commercialbubble area = total parameters · hover for detail

GLM-5.2 posts the highest reported score in this set — 91.2 GPQA-Diamond on ~753B total / ~40B active parameters, MIT, weights on Hugging Face with no regional restrictions aitoolsreview.co.uk

Four models cluster at 90+ GPQA Diamond — GLM-5.2 (91.2), Kimi K2.6 (90.5), Hy3 (90.4) and DeepSeek V4 Pro (90.1) — three of the four under MIT-family licenses ai-beat.github.io

Efficiency outlier: Qwen3.6-35B-A3B reaches 86.0 GPQA while activating only 3B parameters per token and fitting in ~20 GB locally, Apache 2.0 ai-tldr.dev

Gemma 4 is the first Gemma generation under fully permissive Apache 2.0, replacing the Gemma Terms of Use — its 31B dense flagship scores 84.3 GPQA Diamond hosn.om

All nine qualifying models

Model	Total params (active)	Release date	License	Reported GPQA	Sources

Method: evidence table of 13 rows compiled from public web sources, retrieved as of 2026-09-02. Three Gemma 4 rows and two DeepSeek V4 Pro rows were consolidated into one canonical row each (DeepSeek V4 Pro dated to its first public release, 2026-04-24); DeepSeek V4 Flash was excluded for lacking a GPQA score. "More than 30B" counts total, not active, parameters; "after January 2026" means later than 2026-01-31. Permissive includes MIT, Apache 2.0, Modified MIT where commercially usable, and Mistral's reported permissive research-and-commercial license. GPQA scores are shown as reported (usually GPQA Diamond); variants and evaluation settings may differ, so scores are not strictly comparable. Single-source findings are flagged in orange.

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.