AI Daily · 2026-10-18

Tencent Hunyuan released Hy-Memory-30B-A3B-FP8, a quantized version of its ~3B active-parameter long-term memory writer optimized for vLLM, and it edges out general-purpose LLMs such as Claude Opus 5 and DeepSeek V4.1 Flash on a fixed evaluation across memory frameworks. On hardware, NVIDIA's RTX Spark laptops are now available starting at $2,399, bringing Blackwell GPUs, up to 128GB unified memory, and the full NVIDIA AI platform into thin Windows laptops for local AI, creation, and gaming. Simon Willison also highlighted a small front-end trick using document.currentScript to replace inline scripts with rendered content. Overall, the day leans toward on-device AI and memory infrastructure, with a lightweight developer tip alongside.

East Asia · First-hand

Tencent Hunyuan

⭐⭐ [Model Release] tencent/Hy-Memory-30B-A3B-FP8

Tencent Hunyuan released Hy-Memory-30B-A3B-FP8, an FP8-quantized version of Hy-Memory-30B-A3B optimized for vLLM deployment and published under apache-2.0. Hy-Memory-30B-A3B is a compact long-term-memory writer model with roughly 3B active parameters that executes the write path inside external memory frameworks, directly replacing the general-purpose LLM in Mem0, MemOS and Hy-Memory (covering Chat, Working and Coding scenarios) and deciding what information to retain, how precisely to represent it, and how new evidence should update the persistent memory bank. It is trained in two stages, Memory On-Policy Distillation (OPD) followed by Attributed Memory Reinforcement Learning, and transfers directly to unseen memory frameworks such as EverMemOS. In evaluation only the memory writer is changed, with harness version, prompts, retrieval configuration, answer model and judge held fixed, and every benchmark is run on a fixed 20% subsample; each harness score is the equal-weight average of six metrics and the overall score the equal-weight average of four harnesses. Hy-Memory-30B-A3B scores 55.49 overall, ahead of Claude Opus 5 at 54.62, DeepSeek V4.1 Flash at 53.64 and Qwen3.6-35B-A3B at 50.37.

Ecosystem & Beyond (Products / Agents / Tools / Opinions)

Product Update

⭐⭐ [Product Update] RTX Spark Ignites Global Celebration as New Windows PCs Arrive

NVIDIA announced that RTX Spark laptops are now available, bringing Blackwell RTX graphics, up to 128GB of unified memory, and the full NVIDIA AI platform to slim premium Windows PCs. Starting at $2,399, machines from ASUS, Dell, HP, Lenovo, Microsoft, and MSI offer local AI, content creation, and gaming capabilities. The source text is truncated at this point and provides no further details.

Other

⭐⭐ [Other] Inline rendering with document.currentScript

Simon Willison's link blog highlights a trick from Lean Rada that uses the document.currentScript property: you can drop in a tiny inline <script> element that replaces itself when the page loads. For example, "The year is <script> document.currentScript.replaceWith(new Date().getFullYear())</script>" renders the current year. Defining a shortcut function such as $=(...n)=>document.currentScript.replaceWith(...n) yields a tiny template language, e.g. "The year is <script>$(new Date().getFullYear())</script>".