Unverified50% confidenceFactExact time
On the MLX side, upstream mlx-lm does not yet recognize hy_v3, but hermitdave's Mixed4-6 build bundles hy_v3.py and points to it via config, so it can be loaded directly with stock mlx-lm; the uploader reports decoding at about 65 tok/s on an M3 Max 64 GB, perplexity within 0.94% of the 8-bit version, and 70/70 factual recall on the extraction test.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedThe Qwen 9B model has a peak memory usage of approximately 5.1 GB when loaded in MLX format.70% similarUnverifiedIn a fair 4-bit comparison, MLX with D-Flash 2 speculative decoding ran at 16.6 tokens/s, versus Ollama Q4's 5.7 tokens/s.69% similarUnverifiedThe Q4_K_M/GGUF version of Qwen 27B decodes at approximately 10.7 tok/s on the same machine, roughly 1/3.8 of the MLX version.69% similarUnverifiedOn an M4 24GB device, a 9B model (such as qwen3.5-32k:9b-mlx) runs fast with low memory usage, but is unable to effectively identify code issues.69% similarUnverifiedMiniMax H3's weights total 67 GB, comprising a 34 GB DiT backbone, a 27 GB Qwen3-VL encoder, plus VAE66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054170API
curl https://kongchang.com/api/v1/knowledge/claims/1054170MCP
get_claim(id=1054170)