AFM 3 splits into 20B on-device sparse (NAND-to-DRAM weight loading) and Cloud Pro on Google Cloud NVIDIA GPU. Three Google contexts, Foundation Models API opening, and what's still unreleased.
Tested LFM2.5-1.2B-JP-202606 on M1 Max 64GB. llama.cpp Q4_K_M: 208 tok/s decode, JSON intact, model name hallucinated (LFM→FDM). Q8_0: 157 tok/s, no hallucination. Tool calls broken via GGUF.
35M linear projection replaces E4B's 150M 16-layer Vision Encoder. Bidirectional attention in the 48-layer LLM absorbs patch features. Comparison with Fuyu, EVE, EVEv2, and Mono-InternVL.
Hands-on with Tencent Hy-MT2 1.8B Q4_K_M (1.08GB) on M1 Max 64GB via llama-server. JSON, SRT, HTML, glossary, and minority-language prompts with full input-output pairs. The 1.25bit 440MB build does not load on stock llama.cpp 8990, and 30B-A3B (hy_v3) is not in the Mac route yet.
xAI x-algorithm second commit: Phoenix retrieval+ranking runs locally on 537k sports posts with 3GB artifacts. Ad blending and candidate isolation code added since January.
Verdict on GTIG's May 11, 2026 report: the first confirmed AI-generated zero-day, a Python 2FA bypass for an OSS admin tool, was caught by a hallucinated CVSS score and textbook Pythonic code structure.
oMLX 0.3.9.dev2 release notes from the angle of Codex/Copilot on Mac local LLMs: Gemma 4 VLM MTP, DFlash, omlx launch copilot, SSD KV cache — what each changes for agent workflows.
Out-of-bounds read in Ollama's GGUF loader before 0.17.1. If your Ollama API is network-accessible, a crafted model file can exfiltrate env vars, API keys, system prompts, and conversation fragments from process memory.
Checked Fortress Token Optimizer's DEV article and npm/PyPI packages. Polite filler words shrink 11-22%, but running it blindly on system prompts or RAG context can strip constraints that control model output.
Tested Gemma 4 MTP drafter on M1 Max 64GB with mlx-vlm 0.5.0. Only the 26B A4B MoE got +13%; 31B Dense and E4B got slower. Code gen vs short haiku prompts flip the result.
