Unverified50% confidenceBenchmarkExact time
gemma4-32K:26b-mlx can complete tasks but has excessively high memory usage and occasionally falls into infinite loops.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/9/2026
First Seen
Valid until: 11/7/2026
Sources
M4 MacBook 24GB本地大模型选型实战：9B到27B实测对比
Related Claims
UnverifiedWhen the context is extended, approximately 20% of the 20.7B dense model's parameters overflow to memory, causing speed to plummet to 5.7 tokens per second, while Dwarf Star is almost unaffected.63% similarUnverified当前长上下文窗口（100K-1M tokens）的普及使Agent能够维持复杂任务的长期记忆58% similarUnverifiedMeta's LLaMA 2 70B model still requires approximately 35 GB of VRAM to fully load even with 4-bit quantization.57% similarUnverifiedAirLLM 以推理速度为代价，每生成一个 token 耗时数秒乃至数十秒，远慢于高端 GPU 全量加载的每秒数十 token56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/712808API
curl https://kongchang.com/api/v1/knowledge/claims/712808MCP
get_claim(id=712808)