--- title: "AI engineering, October 18, 2026" description: "The Forward Pass daily issue, October 18, 2026." canonical: https://theforwardpass.net/archive/daily/2026-10-18 updated: 2026-10-18 --- # AI engineering, October 18, 2026 The Forward Pass daily issue, October 18, 2026. Source: https://theforwardpass.net/archive/daily/2026-10-18 > This issue is researched and written by AI models, and every fact is checked against its cited source. No human edits it before it is sent. _DeepSeek speeds up Flash-1017, Tencent open-sources Hy Memory 2.0, NVIDIA ships RTX Spark laptops, Amp adds desktop control for macOS and Linux_ You can now buy a Windows laptop with up to 128GB of unified memory. NVIDIA's RTX Spark puts a Blackwell RTX GPU and a Grace CPU on one superchip. Meanwhile, DeepSeek-V4.1-Flash-1017 runs faster and costs less than DeepSeek-V4.1-Flash. Tencent Hunyuan open-sourced Hy Memory 2.0 with an SDK, a 30B-A3B memory model and the MemHub local app. If you read one thing today: DeepSeek-V4.1-Flash-1017. Its MIT-licensed weights let you test the coding and agent improvements locally. ## DeepSeek makes V4.1-Flash-1017 faster and cheaper than DeepSeek-V4.1-Flash, hitting 42.3 on Terminal-Bench **Top Model** · 1,057 Likes If you're choosing a model for agent runs, DeepSeek positions V4.1-Flash-1017 as faster and less costly to use than DeepSeek-V4.1-Flash. This minor update also produces shorter outputs and improves reasoning, coding and visual benchmark results. Here's what changed: - Terminal-Bench 4.0 rises to 42.3 from 31.2 for DS-V4.1-Flash. - DeepSeek reports 77.0 on DeepSWE v1.1, versus 74.2 for DS-V4.1-Flash and 74.0 for Sol Opus 5. - Automation-Bench v1.0.6 reaches 60.5, up from 54.8 for DS-V4.1-Flash. - The MIT-licensed model and weights support local inference, with a 1M-token context window. One catch: There is no Jinja-format chat template. The provided deepseek-recipe toolkit handles request encoding and output parsing, but leaves inference, tool execution and HTTP transport to you. Try it: Use `deepseek-ai/DeepSeek-V4.1-Flash-1017` and the repository's `inference` folder for weight conversion and local inference. Also available via Vercel Sources: [Hugging Face](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) ## Tencent open-sources Hy Memory 2.0 with a 30B-A3B memory model **Top News** Hy Memory 2.0 is Tencent Hunyuan's open-source package for adding working and code memory to AI agents. The release brings the 30B-A3B memory model, an SDK and the MemHub local app together, giving you an open-source package to evaluate for your agent's working and code memory layer. Here's what you get: - The memory model is 30B-A3B and ships as part of the open-source release. - The SDK comes alongside MemHub, a local app included in Hy Memory 2.0. - Tencent reports reducing tokens per question from 1,159 to 703. - Tencent reports a 72.6% result on LoCoMo. Why care? Tencent reports a lower token count per question, and Hy Memory 2.0 gives you the SDK, model and local app in an open-source release. Sources: [pandaily.com](https://pandaily.com/tencent-hunyuan-hy-memory-2-0-open-source-working-code-memory-memhub-apache-2) ## NVIDIA ships RTX Spark Windows laptops starting at $2,399 **Top News** · Official · nvidia.com RTX Spark is a new generation of Windows laptops built around a superchip that combines a Blackwell RTX GPU with a Grace CPU. NVIDIA says laptops are available now starting at $2,399, giving developers and AI enthusiasts a Windows option for running local AI. Here's what you get: - ASUS, Dell, HP, Lenovo, Microsoft and MSI have models available now, with additional systems on the way. - The superchip offers up to 1 petaflop of AI compute. - Unified memory reaches up to 128GB. - NVIDIA provides tested recipes, playbooks and resources to help you run local AI on RTX Spark. Why care? RTX Spark puts NVIDIA's Blackwell RTX GPU and Grace CPU in Windows laptops you can buy now, starting at $2,399. Sources: [blogs.nvidia.com](https://blogs.nvidia.com/blog/local-ai-rtx-spark-windows-laptops-celebration/) ## Signals 1. [SPACE goes open source as the sandbox platform behind Computer](https://www.perplexity.ai/hub/blog/open-sourcing-space) 2. [Amp adds desktop viewing and control for macOS and Linux runners](https://ampcode.com/news/runner-desktop) 3. [CA-MAMBA multi-agent learner earns 2.8-12x baseline reward on MAMuJoCo at equal budget](https://arxiv.org/abs/2610.21319) 4. [arXiv study presents 0.6B-14B open-weight Wolof language models tested on 63 tasks](https://arxiv.org/abs/2610.20824) 5. [VoiceStudio runs voice cloning and video dubbing locally across 646 languages](https://github.com/debpalash/VoiceStudio) · 99 Stars today 6. [NVIDIA releases Dory1.0-4B-NVFP4, letting one reasoning model scale inference compute without retraining](https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-NVFP4) · 369 Downloads