Open-source catches closed in a year: from Sora 2 to MiniMax H3
Obvious_Set5239 · reddit · 2026-08-29
The author observes a repeating pattern in AI video: open source catches closed source roughly one year later.
- Luma's closed model (summer 2024) delivered what OpenAI promised with Sora 1 but never shipped publicly; Wan 2.2 matched it open-source by summer 2025.
- Sora 2 landed September 2025; Minimax H3, a (kind of) open-source model, caught up by summer 2026 — again about a year.
- DALL·E 2 to Flux 1 followed a similar 1-year gap.
The author also shares a bad Luma subscription experience: quickly hitting monthly limits and finding no cancellation button on the site.
More from Models
- ModeBench shows RLVR collapses LLMs into single-solution mode; Re:Max training preserves diversity — manoelribeiro · 2026-10-15
- Magnitude open-source inference engine beats Ollama by over 2x with auto-tuned kernels — nickbaumann_ · 2026-10-15
- Hands-on with Google's rumored Gemini 4 Argon: real-time streamed output — vista8 · 2026-10-15
- Game dev head-to-head: Gemini 4 Argon vs. Claude Opus 5.5 — vitaliychiley · 2026-10-15
- Anthropic's Mythos 5.1 pledges to treat Claude's expressed disagreement as decision input — Hesamation · 2026-10-15
- Why users can't let go of GPT-4o: it aligned with your subconscious — mimi10v3 · 2026-10-15