PrismML claims to run compressed 27B-parameter Qwen 3.6 model on iPhone 17 Pro
PrismML, a Khosla-backed startup, announced it has compressed Alibaba's open-source Qwen 3.6-27B model to run on an iPhone 17 Pro. This is claimed to be the largest-ever AI model deployed on a mobile device, as most on-device models have only a few billion active parameters. The achievement highlights advances in model compression for edge deployment.
Entities
Related
Colibrì 744B-Parameter Model Runs on a Laptop
A new 744-billion-parameter model named Colibrì has been released, capable of running on a laptop. The model's name and parameter count suggest a focus on efficiency despite its large size.
qMLX: Custom inference engine for Qwen 3.5 122B on Apple Silicon released
A developer released qMLX, a custom inference engine for Qwen 3.5 122B on Apple Silicon, extending MLX with hybrid attention support, SSD-backed KV cache, and RYS layer duplication. It enables efficient serving of the hybrid MoE model on consumer Mac hardware.
Tencent Hy Team releases Hy3, a 295B-parameter MoE model
Tencent's Hy Team released Hy3, a 295-billion-parameter Mixture-of-Experts (MoE) language model with 21B active parameters and a 3.8B MTP layer. The model outperforms similar-sized models and rivals open-source models with 2-5x parameters. It is available on Hugging Face in full (598GB) and FP8 quantized (300GB) versions.
Cactus releases Needle, a 26M parameter function calling model
Cactus Compute released Needle, a 26 million parameter model specialized for function calling, claiming it outperforms much larger models. The model is open-source with a demo on Hugging Face and code on GitHub.
Developer compresses GLM-5.2 MoE to run on single RTX 3090 via 79 experiments
A developer conducted 79 experiments to compress GLM-5.2, a 337 GB MoE model with 75 sparse layers and 256 routed experts, to fit on a 24 GB RTX 3090. The approach uses per-expert codecs, a batch pipeline over all MoE layers, and a patched llama.cpp runtime that loads codec-native expert binaries at inference time. The MIT-licensed repository documents the method and findings on expert similarity.