Short Stack #1Wed, Oct 14, 2026
01 Vibe coders
Gemini 4 Argon is out
Google shipped Gemini 4 Argon on Oct 13. It's Google's new frontier model for long, multi-step coding work like migrations. The API price starts at $2/M input and $10/M output tokens, and the input+output limit is 1M tokens. It's on the paid Gemini API tier, in AI Studio and Android Studio, and in Antigravity for AI Ultra subscribers.
⏱ 10-MIN TRYPick a messy file in a side project. Paste it into AI Studio with Argon and ask for a refactor plan. Then compare that plan with your usual model's.
source ↗ blog.google02 Agent devs
ScaleDown's single-task models for agent loops
ScaleDown launched Constrained Decoding Models on Oct 13: Summarize, Compress, Extract and Classify. Each one only returns the answers its task allows, and each takes up to 1M input tokens. LangChain shipped `langchain-scaledown` middleware the same day. Extract and Summarize cost $0.04/M input tokens on OpenRouter. The speed and cost numbers come from the vendor and haven't been checked independently.
⏱ 10-MIN TRYRun `pip install langchain-scaledown`. Add the tool-result compression middleware to one Deep Agent, then compare token counts before and after.
source ↗ aicoder.com03 Designers
Canva's AI Connector, powered by MCP
Canva's AI Connector, which runs on its MCP server, lets ChatGPT, Claude, Gemini and Copilot create, search and edit on-brand Canva designs using your brand kit. The results stay fully editable.
⏱ 10-MIN TRYConnect Canva in Claude or ChatGPT settings, or type `@canva` in Gemini. Then ask for a 3-slide deck that uses your brand kit.
source ↗ canva.com04 Non-technical founders
A cheaper frontier model you can use in a chat
You don't need the API to use Argon. Google AI Pro and Ultra subscribers can use it in the Gemini app and in AI Mode in Search today.
⏱ 10-MIN TRYGive Argon your landing-page copy and your top 3 competitors' URLs. Ask for a positioning gap analysis and 3 headline rewrites.
source ↗ blog.google05 Wildcard
NVIDIA open-sources Run:ai
On Oct 13 NVIDIA released the full Run:ai GPU orchestration platform as open source. That includes KAI Scheduler and a GPU Fractioning Operator, which you can also adopt on their own.
⏱ 10-MIN TRYRead the KAI Scheduler README and see whether GPU fractioning could let you run 2 small models on 1 GPU.
source ↗ developer.nvidia.com