Four companies post-train their own video models after MiniMax H3 goes open source
videorouter · reddit · 2026-10-31
After MiniMax open-sourced H3, four companies have already post-trained it into their own video models. The leaderboard shows MiniMax H3 at 1277 (variants up to 1286), competing against closed models like Pika H3 Flex (1293), Seedance and Wan 3.0, with H3 among the few open-weight options — open foundations are rapidly closing the gap with closed video models.
More from Venture
- Fear & Greed Index plunges to 22 'Extreme Fear', down 13 points in a day — JOBhakdi · 2026-10-31
- Anthropic's S-1: top 5% Claude Code users spend 20x more, NRR near 900% — HarperSCarroll · 2026-10-31
- Nvidia's enterprise/neocloud segment growing 138.5% could overtake hyperscalers in 2-6 quarters — Beth_Kindig · 2026-10-31
- Anthropic's $2T valuation means ~$377M value per employee, signaling output-org decoupling — signulll · 2026-10-31
- Survey of 200 CIOs/CTOs: enterprise AI ROI concentrates by workload, rises with scale — BenBajarin · 2026-10-31
- Jinxo Gives Away 3 Months of Claude Max ($600) With Its $20 Plan — gaganghotra_ · 2026-10-31