Not all open weights are equal: comparing release packages across labs
WarmMycologist9549 · reddit · 2026-11-20
With Tencent's Hy4 weights now out, the author compares recent open-weight releases from Chinese and international labs and finds the packages differ widely:
- Google Gemma 4: 2B–31B range under Apache 2.0.
- OpenAI gpt-oss: 20B and 120B under Apache 2.0, with a separate usage policy.
- Meta Llama 4 Scout: 109B total / 17B active params, Llama Community License.
- DeepSeek V4.1 Flash 1017: MIT license.
- Tencent Hy4 / Hy4 Flash: Tencent Hy License.
- Qwen3.8 family: mixed licenses across models, from Apache 2.0 to Qwen's own terms.
The core point: open weights are only part of the picture. License terms, checkpoint availability, and access to training materials all determine what developers can actually do, and these dimensions deserve scrutiny when evaluating any "open" release.
More from Models
- Nathan Lambert tries Gemini Neon in Antigravity: feels in distribution of Claude Code and Codex — natolambert · 2026-11-20
- Ulms ERT training makes vision nets far easier for LLMs to edit in weight space — Timo Ropinski · 2026-11-20
- Polymarket odds say Gemini will top Artificial Analysis leaderboard by end of November — LeviTurk · 2026-11-20
- Vercel: decision models hit 26% of gateway requests; Anthropic wins new teams 2:1 — airesearch12 · 2026-11-20
- Leaked Gemini 4 lineage: neon scales hydrogen's base, beats carbon at lower cost — haider1 · 2026-11-20
- Hobbyist Converts Qwen3.5-2B Dense to MoE, With Full Benchmarks — Bayliner1980 · 2026-11-20