Agentic workloads get own CPU; Apple puts frontier models on desks
Mac Studio M5 Ultra, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Apple
Mac Studio M5 Ultra
DEPLOYif you run large LLMs on device
WHAT SHIPPEDMac Studio with M5 Ultra scales to a 36-core CPU, up to an 80-core GPU, and 512GB of unified memory with 1.2TB/s bandwidth, delivering up to 4.3x the peak AI compute performance of M3 Ultra and 9.8x more than M1 Ultra.
THE CATCHA cluster of four Mac Studio systems delivers up to 3x faster AI inference than a single system using Thunderbolt 5 and RDMA, creating a shared memory pool that loads the largest frontier-class open-weight models available today.
IF YOU RUNlocal AI models in LM Studio Bionic or need to run enormous LLMs entirely on device with complete privacy
SOURCEApple, Apple introduces new Mac Studio with M5 Max and M5 Ultra
The Deploy Log
Models and APIs
| DEPLOYif you use Kiro for AI-native coding GPT-5.6 in Kiro. The GPT-5.6 model family, including Sol, Terra, and Luna, is now available in Kiro, and testing found that on Terminal-Bench 2.1, GPT-5.6 Terra completed successful tasks in Kiro at roughly 82% cost reduction. OpenAI |
| WATCHif you run an academic or nonprofit lab Claude team plan for scientists. Anthropic is opening 10,000 seats for scientists to access Claude subscriptions free and at discounted rates for one year, with standard seats free and premium seats with 5x usage limits at $15 per month. Anthropic |
Tools and products
| DEPLOYif you manage Vercel projects with coding agents Vercel Security Dashboard. The Vercel Security Dashboard is now generally available on all plans, automatically flagging issues like team members without 2FA, long-lived credentials, public preview deployments, and stale environment variables. Vercel |
| DEPLOYif you connect agents to external services Vercel Connect. Vercel Connect is now generally available on all plans and in v0, replacing stored long-lived provider secrets with short-lived scoped tokens requested at runtime, with 100+ preset connectors and pricing at $3 per 1,000 token requests and $0.95 per 1,000 triggers on Pro. Vercel |
| DEPLOYif your sandboxes access regional databases or storage Vercel Sandbox regions. Vercel Sandbox now runs globally starting with four regions: iad1, sfo1, cle1, and cdg1, with region selection on all plans and failover regions configurable for Pro and Enterprise teams. Vercel |
| WATCHif you need long-horizon coding or document analysis Hy4 Preview on AI Gateway. Hy4 Preview from Tencent is now available on AI Gateway, an open-source Mixture-of-Experts model with 770B total parameters and 49B active per token, serving a context window of 1M tokens. Vercel |
| WATCHif you do financial research and analysis Ling 3.0 Flash Fin. Ling 3.0 Flash Fin from Inclusion AI is now available on AI Gateway and free to use through September 25, with a 256K-token context window, up to 32K output tokens, and support for reasoning and function calling. Vercel |
| DEPLOYif you run coding or multi-step agent workflows Qwen 3.8 Flash. Qwen 3.8 Flash from Alibaba is now available on AI Gateway as alibaba/qwen3.8-flash, taking text and images as input with a 1 million token context window and up to 65k tokens in a response. Vercel |
| DEPLOYif you build multimodal coding agents GLM 5.3 Flash. GLM 5.3 Flash from Z.ai is now available on AI Gateway as zai/glm-5.3-flash, a multimodal coding model with a 1M-token context window that accepts text and image inputs and supports function calling, structured output, and streaming. Vercel |
| DEPLOYif you want real-user performance data free Speed Insights free tier. Speed Insights now has a free tier available on every plan for any number of projects, including 10,000 events per team every 30 days, with the paid product renamed Speed Insights Plus. Vercel |
Left out
- Mac mini with M6 and M5 Pro: The page text is truncated before the pricing and availability section completes, so the availability quote cannot be copied exactly from th
- Gemini 3.5 Transcribe: the publisher's page does not show it available yet
- ChatGPT for Teachers expansion: a figure in the draft could not be found on the publisher's own page
- Admin plugin for ChatGPT Work: the publisher's page does not show it available yet
- Claude Managed Agents with Chat SDK: the publisher's page does not show it available yet
- langchain.mcp adapter: the publisher's page does not show it available yet
- Notion agent suggested edits: a figure in the draft could not be found on the publisher's own page
- Gemini 3.5 Transcribe on AI Gateway: the publisher's page does not show it available yet
- Muse Image on AI Gateway: the condition it attaches to the call did not hold up
- BotBase for Operators: it did not carry a number that moved
- CSE preview in Drive: it did not carry a number that moved
- eve agents from dashboard: it did not carry a number that moved
- Gemini Drive classification: it did not carry a number that moved
- NVIDIA Vera CPU: the announcement hedges, so nothing has deployed yet
The pattern
The same week Apple ships a 512GB unified memory Mac Studio that loads frontier-class open-weight models on a desk, Vercel makes its security and connectivity layer generally available on all plans and floods AI Gateway with four new models, including a 1M-token Hy4 Preview and a free Ling 3.0 Flash Fin through September 25. The hardware move and the platform move are one motion: agentic workloads now get a local CPU with 1.2TB/s bandwidth while hosted inference drops to $3 per 1,000 token requests and an 82% cost reduction on Terminal-Bench 2.1.
Anthropic opens 10,000 seats for scientists at $15 per month for premium access, and Vercel gives every team 10,000 Speed Insights events per 30 days. The pattern is not cheaper compute alone. It is distribution of frontier and near-frontier models across four sandbox regions, iad1, sfo1, cle1, and cdg1, with short-lived scoped tokens replacing long-lived credentials, so a 36-core CPU and a 770B-parameter model sit inside the same security boundary.