// 6 期 · 标注日期、可引用

本地AI报告

关于本地LLM和硬件现状的双周刊——新模型、GPU价格变动、工具更新。有日期、可引用、有存档。

通过邮件接收每一期Local AI Digest 每两周发送一次。无垃圾邮件，随时可退订。订阅简报 →
最新 · 报告Local AI Report #6 — Best Local Coding Models: Three Picks for Every Hardware TierThe best local coding models for 8–16 GB, 24–32 GB and 128 GB machines: three picks per tier, what they run on, and how far behind Claude they score.阅读本期 →
  • Entry level (8–16 GB): K2 Horizon 7B. A 5.6 GB download that scores 21 on the independent Artificial Analysis index — nearly double Qwen3.5-9B's 11. Runners-up: Qwen3.5-9B (the easiest to install) and K2 Horizon 3.7B for 8 GB.
  • Mid-range (24–32 GB): Qwen3.8-27B. Index 34, an 18 GB download, and about 21.9 GB at 64k context. Runners-up: K2 Horizon MoVA 36B-A4B (25, on a 32 GB card) and the dedicated coder Devstral Small 2.
  • High-end (128 GB): Qwen3.8-Flash-Next. Index 40, a 120 GB download — tight on a DGX Spark or Strix Halo, too big for a 128 GB Mac. Runners-up: DeepSeek V4 Flash 0731 at 3-bit and Qwen3.8-27B at full precision.
  • The gap to the subscriptions is real. The best model on one GPU scores 34; Claude Opus 5.5 scores 58 (index v4.3.2, checked 2026-10-08). That is the budget subscription tier, not the flagship.
报告 · 2026年9月14日Local AI Report #5 — Europe's AI Compute Gap and SovereigntyThe EU runs just 2 GW of AI compute to America's 35 GW. Inside Europe's 19 AI Factories, the gigafactory tender, and Luxembourg's 20-exaflop sovereignty bet.europesovereigntyinfrastructure报告 · 2026年8月31日Local AI Report #4 — Best Laptops for Local AIIn a laptop, memory bandwidth and memory capacity sit at opposite ends of the price list. The fastest memory you can buy comes in the smallest quantity, the largest comes on the slowest bus, and exactly one part escapes the trade-off.LaptopsHardwareApple Silicon报告 · 2026年8月25日Local AI Report #3 — Best Small LLMs for 8 GB and 16 GB RAM LaptopsAn 8 GB laptop leaves about 6.4 GB for a model, and a 16 GB laptop about 12.8 GB. Here is what actually fits, how much context you get, and how fast it runs with no discrete GPU.LaptopsSystem RAMCPU Inference报告 #2 · 2026年7月8日Local AI Report #2 — The Best Open-Source Coding Models Right NowA field guide to the strongest open-weight coding models in mid-2026: the SWE-bench frontier (DeepSeek V4-Pro, GLM-5.2, Kimi K2.6), the best coder you can actually download (Qwen3-Coder), and the setup that gives the most code per dollar on a single GPU.DigestCodingModels报告 #1 · 2026年7月7日Local AI Report #1 — The Mid-2026 Local LLM & Hardware LandscapeThe first biweekly digest on the state of local LLMs: the mid-2026 model generation (Gemma 4, Qwen 3.6, DeepSeek V4), where GPU prices actually stand, and what the current best-value setup looks like.DigestModelsHardware