“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 broad live-web scan as of 5 September 2026 found 35 qualifying release records, consolidating to 32 distinct models and checkpoints once three naming aliases are merged. Every model here carries an Apache 2.0, MIT, or modified-MIT license, more than 30 billion total parameters, a release date between 1 February and 5 September 2026, and a reported GPQA figure (almost always GPQA Diamond, unnormalized across evaluation settings).
| Model | Parameters | Release date | License | Reported GPQA | Developer | Source |
|---|
Method: broad indexed-web scan as of 2026-09-05; 35 qualifying release records consolidated to 32 after merging three naming aliases (Qwen 3.5 → Qwen3.5-397B-A17B; Gemma 4 31B variant → Gemma 4 31B Dense; Tencent Hy3 295B-A21B → Hunyuan 3), keeping the more specific row with more supporting pages. Criteria: Apache-2.0, MIT, or source-explicit modified-MIT license; more than 30B total (not active) parameters; release 2026-02-01 to 2026-09-05; a reported GPQA figure, overwhelmingly GPQA Diamond, not normalized across evaluation settings. Dated, GA, and quantized checkpoints stay separate when they are distinct releases. A five-row first-party cross-check of official pages was used only for corroboration. “All” means all qualifying models found in this scan, not a guarantee against unindexed releases; rows backed by one indexed page are flagged.