ExplainersMiniMax H3 guide: specs, license and prices, updated weeklyopen weights
Every SLOP TV News story about open weights.
ExplainersMiniMax H3 guide: specs, license and prices, updated weekly
NewsLiquid AI's LFM3 reads text, video and audio on a phone
OpinionPika H3 Flex ships open weights under a license that bars the US
Pika H3 Flex
Pika H3 Flex cuts video generation from 50 steps to six
Pika's post-trained MiniMax H3 turns out a 15-second 768p clip in as little as seven seconds of denoising time, and the API starts at a cent a second.
Maya 2 Native
Maya 2 Native opens streaming speech in 12 languages
Maya Research published Apache 2.0 weights and inference code on October 25, with 20 voices and cloning from 3 to 10 seconds of speech.
MiniMax H3
MiniMax H3's 2K finish is now an open 3-step pass
The refiner and decoder still carry MiniMax's H3 license, which excludes the United States, the European Union, the United Kingdom and South Korea.
Nvidia
Nvidia's Flora labels robot video for an estimated 72 cents an hour
The open suite of three fine-tuned Qwen models undercuts GPT-6 Astra by more than a hundredfold on Nvidia's own cost estimate.
open weights
Open weights is becoming a license tier, priced by revenue
ComfyUI is paying $25,000 for open-source rebuilds of Higgsfield's apps while Nvidia, whose cards are the prizes, backs Higgsfield.
EdenLive
EdenLive renders 720p video at 24.7 FPS on one RTX 5090
The open-weight renderer's code and main weights are Apache 2.0, but the streaming service needs an 80 GB NVIDIA GPU and CUDA 13.
UniAVC
UniAVC edits a video and its soundtrack from one instruction
HKUST and Huawei Research released the weights and code, but peak memory is about 39 GB even with offloading, and there is no hosted demo.
LTX-2.5
LTX-2.5 ships as a 22B open-weight video model with audio
One prompt can now produce a sequence of shots instead of a single take, and the model card publishes no benchmark scores.
Hy-Nomas
Tencent's Hy-Nomas world model ships with no published scores
The Apache-2.0 weights, code and a 280-case benchmark went up on October 23 with no paper and no results, and one configuration targets a single RTX 4090.
nvidia
Nvidia's $4,999 DGX Spark runs some AI video models, not all
NVIDIA's six OEM partners start selling the 64GB box on October 23, and two units linked by a QSFP cable pool to 128GB for models up to 200 billion parameters.
matrix-game
Matrix-Game 4.0 renders game worlds at 60 FPS on one RTX 5090
Skywork AI's open-weights release splits an executable game engine from a causal video model and ships three playable scenes under Apache-2.0.
UniAVC
UniAVC edits a video and its soundtrack from one prompt
The 19B-parameter model is free to download under the LTX-2 Community License, but it needs about 39 GB of VRAM to run, which rules out consumer GPUs.
EdenLive
VAST open-sources EdenLive, a real-time video renderer
The 5B model turns rough game footage into detailed frames at more than 24 frames per second on a single RTX 5090; the 14B build runs at 480p.
minimax-h3
MiniMax H3's open license excludes the US, UK and EU
Builds keep appearing under it, including one from NVIDIA Research, while a rival open video model ships with no territory clause at all.
FLUX 3 Image Klein
FLUX 3 Image Klein ships open weights for 8 GB GPUs
Black Forest Labs releases the 5.6-billion-parameter image model under a license that allows free commercial output below $5 million in annual revenue.
FastH3-Omni
FastH3-Omni adds reference images and audio to MiniMax H3
The open-weight model previews on the Reactor API through October 20, when FastVideo says the final weights and training code ship.
Agnes-Video-2.0-MoE
Agnes Video 2.0 MoE ships open weights under Apache 2.0
The 28.7-billion-parameter model generates video with audio, but its Apache 2.0 label stops at the Agnes core weights and excludes the text encoder and VAE.