Coding agents
VS Code 1.143, GPT-6 on Google Cloud, NVIDIA DSX POD: the October 22 roundup
VS Code 1.143 pits several models against one task, Google Cloud adds GPT-6 to Model Garden and NVIDIA promises AI factories set up in 14 days.
Three announcements from October 19 to 21, 2026, all taken from official sources: the editor, the cloud and the hardware are moving at the same time around coding agents.
VS Code 1.143: Copilot harness by default and a model face-off (Oct 19)
Microsoft released Visual Studio Code 1.143 (release notes).
- Copilot harness by default: new agent sessions run on the harness built on the Copilot SDK, available from both the editor and the Agents window.
- Compare models (preview): one prompt starts several attempts in parallel, mixing Copilot, Claude and Codex harnesses and reasoning levels. An optional "Judge" model scores the attempts (time, tokens, rationale), then a "Synthesize" action combines ideas on request. Setting:
sessions.chat.compareAgents.enabled. - On-device model (experimental): MAI Code 1.1 Flash runs locally through llama.cpp; those responses do not use AI credits. It needs GitHub sign-in and an NVIDIA GPU with at least 96 GB of memory.
- Large chats: per Microsoft's measurements, a cold open of a chat with about 1,150 requests drops from 35.6 s to 9.0 s.
- Copilot CLI: the VS Code installer now offers to install it (Windows, Linux, macOS).
GPT-6 lands in Google Cloud's Model Garden (Oct 21)
Google Cloud is adding GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna to Model Garden, in private preview for select enterprises (Google Cloud blog).
- Billing: the models are sold through Google Cloud Marketplace, show up on the Google Cloud invoice and draw down existing spend commitments.
- Positioning: Google says it is the only hyperscaler offering all three leading closed model families (Gemini, Claude, GPT), among more than 200 models.
- Codex and Antigravity: OpenAI announces its Codex workflows in Model Garden and its models "soon" in Google Antigravity; general availability is planned "in the coming weeks". OpenAI adds that Microsoft remains its primary cloud partner (OpenAI).
- What is missing: Google has published no Model Garden prices, regions or firm date; zero data retention and Confidential Compute are listed as "coming soon".
NVIDIA DSX POD: an AI factory delivered in containers (Oct 21)
At GTC Berlin, NVIDIA introduced DSX POD, a reference design for Vera Rubin AI factories assembled and tested at the factory inside containers (NVIDIA Newsroom).
- The figure: at an already prepared site, deployment would drop from six months to as little as 14 days, according to NVIDIA. Site preparation is not counted.
- Partners: Bull, Dell, HPE, Lenovo, Schneider Electric, Vertiv, Foxconn and Crusoe, among others.
- Timing: first deliveries in early 2027.
What it changes
For developers, VS Code turns model comparison into an editor feature, putting Claude and Codex in the same window as Copilot. For enterprises, the choice of model is coming apart from the choice of cloud. For infrastructure, NVIDIA is now selling time to service as much as the chip.
Sources
- Microsoft: VS Code 1.143 release notes (verified 2026-10-22)
- Google Cloud: Google Cloud brings OpenAI models to customers (verified 2026-10-22)
- OpenAI: OpenAI models are coming to Google Cloud (verified 2026-10-22)
- NVIDIA: DSX POD AI Factories (verified 2026-10-22)
Spotted an error in this article? Report it · Our corrections policy