Unverified50% confidenceFactExact time
Simon Willison used LM Studio to load the 17GB Qwen 3.8 27B Q4 version and reproduced it on a laptop: Breakout, Pong and Pac-Man were correct, Freeway failed (guessed as Pac-Man), three out of four.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Can Qwen 3.8 27B Really Read Neural Network Weights? A 16GB Hands-On Test
youtubeOpaya11/4/2026
Related Entities
Related Claims
UnverifiedLlama 3 70B量化版（Q4精度）仍需约35GB内存67% similarUnverifiedThe blogger koustubh runs the 4-bit GGUF quantized version (UD-IQ4_NL) of Qwen3.5-122B-A10B on a Mac Studio, actually occupying about 62GB of memory.66% 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.66% similarUnverifiedqwen/qwen3.6-35b-a3b is a MoE (mixture-of-experts) model in the Qwen series, and of its 35B parameters only about 3B are activated per token, so it can run at over 100 tok/s on unified-memory devices such as a Mac Studio.65% similarUnverifiedLoading the Qwen3-VL-8B model alone occupies approximately 17–18 GB of VRAM.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135725API
curl https://kongchang.com/api/v1/knowledge/claims/1135725MCP
get_claim(id=1135725)