Unverified50% confidenceFactExact time
Officially released resources include two base models (BF16 and INT8 quantized versions), three text models (BF16, INT8, and FP8), one BF16 VAE, and two PE models supporting image editing and text-to-image generation respectively.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/24/2026
First Seen
Valid until: 12/23/2026
Sources
Related Entities
Related Claims
UnverifiedYuE2 provides a dedicated FP16 version to cover users with Turing architecture GPUs that do not support BF1662% similarUnverifiedThis paper adopts an INT8+bf16 mixed-precision strategy: the original model weights use INT8 quantization (via the LLM.int8() method from the bitsandbytes library), while the LoRA components use bf16 with fp32 master weights.61% similarUnverifiedLTX 2.5 splits model components into separate files and directly provides INT8 and NVFP4 quantization formats.61% similarUnverifiedQwen3.8 27B officially provides a BFloat16 full-precision version and an FP8 version; Unsloth has released a 4-bit version with NVFP4 quantization.60% similarUnverifiedLocal deployment of MiniMax H3 requires downloading four types of model files: the Diffusion Models base model (approximately 19GB), text encoders, 4-step Lora, and two VAE models.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/947461API
curl https://kongchang.com/api/v1/knowledge/claims/947461MCP
get_claim(id=947461)