Tested on M1 Max 64GB: hooked Qwen3.6-35B-A3B's MoE router in mlx-lm, pre-warmed the top-20 hot experts, still ~62 tok/s vs plain mmap cache. Plus the Metal OOM on Qwen3.5-122B.
Kimi K3 activates 16 of 896 experts per MoE layer. From the July 27 tech report: experts are sliced FFNs, not specialists, and the router is one linear projection.
Qwen3.8 Max preview tested via ambassador early access: OpenAI-compatible API behavior, reasoning_tokens, always-on thinking, and the unknowns behind the 2.4T parameter count.
Qwen's Gated Attention (NeurIPS 2025 Best Paper) puts a per-head sigmoid gate on SDPA output. First-token attention drops 46.7%→4.8%, max activation 1053→94. Why it works and how Qwen3-Next uses it.
Kimi K3 is API-only for now: a 2.8T MoE with 1M context via Kimi Delta Attention, open weights promised by July 27 under Modified MIT, and a reasoning-token overhead worth pricing in.
DeepSeek-V4-Pro-DSpark isn't a new base model. It's the same 1.6T V4-Pro checkpoint plus a DSpark speculative-decoding head (~893GB). What config.json and the DeepSpec repo reveal, and why there's no speed benchmark yet.
After a US order pulled Claude Fable 5, which Chinese models drop into Claude Code? Kimi K2.7 Code, Qwen3.7 Max, DeepSeek V4 and GLM-5.1 — constraints, VRAM, benchmark caveats.
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.
After Xiaomi MiMo-V2.5's weights went public, I checked whether it runs on Mac/ROCm or on cloud GPU (RunPod/GCE). It's still rough on local hardware, but RunPod's 4x H200 runs it for ~$14/hr and GCE Spot H100 brings it down to ~$1.6/hr.
Inclusion AI released LLaDA2.0-Uni. A 16B MoE diffusion LLM that handles image understanding, 1024px image generation, image editing, and interleaved text-image generation in a single model.
Hands-on running inclusionAI Ling-flash-2.0 (100B / 6.1B active, MXFP4 quant, 54.7GB) on SwiftLM via mlx-swift-lm on an M1 Max 64GB. Covers bailing_moe + MXFP4 support check in mlx-swift, the startup surprise, and what --stream-experts actually saves.
