Tutorial: Wiring GPT, Claude and Other Models into VS Code Without Copilot

On October 30, 2026, a tutorial published on Juejin walked VS Code developers through connecting external models such as GPT, Claude, DeepSeek and GLM to the editor. The piece targets developers who work in VS Code on projects and would like to call these third-party LLMs while coding, but prefer not to rely on the models shipped with VS Code by default. According to the summary, the article explains how to wire external models into the editor.

The same day brought an unrelated AI release. At 15:31, QbitAI reported that the NVIDIA and MIT team had published the model weights for Open-H3-Regenerate-2K, a video enhancement project, on Hugging Face. An earlier report from Leiphone at 15:06 described the work as coming from the Efficient-Large-Model team, a group of NVIDIA and MIT researchers. The pipeline is built on MiniMax H3 and regenerates low-resolution video drafts into high-resolution output, supporting 2560×1440 2K results and native 4K regeneration up to 3840×2160. Inference code is available in the NVlabs/Sana repository, so developers can deploy the high-resolution enhancement flow themselves.

The Leiphone piece adds context on how MiniMax H3 is structured. The full system spans multimodal context handling, base audio and video generation, and high-resolution regeneration. H3-Base, whose weights are already open, produces 768P audio and video, while the official H3-Regenerate-2K module is served through an API. The newly released project reuses the base model's generative capabilities to offer a self-hostable implementation of the regeneration stage. During fine-tuning, the researchers kept the H3 Transformer backbone and video VAE parameters frozen and trained only a new refinement LoRA. At inference time, that refinement LoRA is merged with community adapters such as LightX2V FL2VA Turbo and LMS Sharp, completing the regeneration through a three-step denoising process.