Fifteen open-weight models over 30B ship permissive licenses and GPQA scores in 2026 — Kimi K3 leads at 93.5

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.”

The screen found 15 canonical model variants released between 2026-02-11 and 2026-07-16, each with an Apache 2.0, MIT, or commercial-use Modified MIT license, strictly more than 30 billion total parameters, and a reported GPQA or GPQA Diamond score. These are 15 qualifying rows from an open-ended ecosystem, not 15 out of a fixed registry; one row, GLM-5.2-NVFP4, is single-sourced to its NVIDIA model card.

Apache 2.0 MIT Modified MIT (commercial use) Filled ring = GPQA Diamond setting · bubble area = total parameters (log)

Kimi K3 posts the highest score, 93.5% on GPQA Diamond, and is also the largest qualifier at 2.8 trillion total parameters under a Modified MIT license · local-ai-zone.github.io

Gemma 4 31B dense is the smallest qualifier at 30.7B total parameters — just over the 30B bar — with 84.3% GPQA under Apache 2.0 · huggingface.co

Mistral Small 4’s primary report of 0.70 is read as 70%; secondary reports cite 71.2% under a GPQA-Diamond setup, so its score is evaluation-dependent · happycapyguide.com

GLM-5’s release date is 2026-02-11 per the Z.ai primary announcement; some secondary reports say 2026-02-12, a one-day discrepancy resolved in favor of the primary source · z.ai

GLM-5.2-NVFP4 is single-sourced: one host, one page, its NVIDIA Hugging Face model card, and its 89.39 GPQA Diamond score is a quantized NVFP4 evaluation · huggingface.co

Model	Developer	Released	Total params	Active	License	GPQA	Source

Click a column header to sort. “GPQA” column notes the benchmark setting: plain GPQA vs GPQA Diamond, tools/settings vary and scores are not always directly comparable.

Method: 15 qualifying model variants as of 2026-09-06, from 610 structured model mentions across 1,483 unique web results plus a first-party/model-card search across 1,850 unique results. Candidates were normalized and kept when backed by at least two independent hosts or a first-party model-card source; GLM-5.2-NVFP4 is single-sourced. “More than 30B” means >30B total parameters; “after January 2026” means release on or after 2026-02-01. Permissive = Apache 2.0, MIT, or commercial-use Modified MIT; custom/community “open” licenses were excluded. Scores are as reported by each source; source URLs span many independent hosts and no single URL supports more than half the rows.

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.