Owner review
This week's radar, before it ships
The low-oversight control room: review detected changes, approve the brief, and keep an eye on subscribers. Reviewing live pipeline items below.
Detected changes — live pipeline queue
17 of 17 included in the next brief. Items are pending your review.
HubSpot ships a refreshed design and HubSpot context inside ChatGPT (September 2026 updates)
September product updates include a refreshed HubSpot design, your HubSpot context now available inside ChatGPT (OpenAI's Small Business Collection), Contracts opened to your wider tech stack, and connector updates for Claude and the remote MCP server.
Next step: Skim the September update list; the ChatGPT context link is worth testing if your team uses both tools.
Midjourney V8.2 is now the default (released July 24, 2026)
V8.2 focuses on aesthetics, image quality, and Personalization, and introduces the new Edit Model, replacing Omni Reference, Character Reference, and the Retexture tool. HD images cost 1.3 GPU-minutes versus 0.8 for SD.
Next step: Expect output style to shift; re-test your go-to prompts before client work.
OpenAI publishes its latest SOC 2 report (covering Jan 1 – Jun 30, 2025)
The report covers Security, Availability, Confidentiality, and Privacy controls for the API Platform, ChatGPT Enterprise, ChatGPT Edu, and ChatGPT Team. Customers with active trust.openai.com accounts can download it under Documents.
Next step: Download it for your compliance file if you rely on OpenAI at work; otherwise no action needed.
OpenAI released GPT-6 Astra; standard pricing $5 per 1M input tokens
GPT-6 Astra is positioned for reasoning, coding, computer use, research, and document creation, carrying tasks from request to finished result. Standard pricing is $5 per 1M input tokens, $0.50 per 1M cached input, $25 per 1M output; billing begins October 5.
Next step: Nothing required; trial it on one real task before switching workflows.
Gemini antigravity-preview-05-2026 shuts down October 5, 2026
Gemini's deprecations page (updated September 15, 2026) lists antigravity-preview-05-2026 with a shutdown date of October 5, 2026.
Next step: Search your apps and scripts for antigravity-preview-05-2026 and move to a supported model before October 5.
Anthropic is retiring Claude Sonnet 4.5 — Sonnet 5.5 is the replacement
Anthropic announced the deprecation of claude-sonnet-4-5-20250929, with retirement on the Claude API. On September 28, 2026 it launched Claude Sonnet 5.5 across the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.
Next step: Replace claude-sonnet-4-5-20250929 with the Sonnet 5.5 model ID wherever it is pinned.
Gemini cache storage price doubles on January 1, 2027
Context-caching storage is $0.50 per 1M tokens per hour through December 31, 2026, then $1.00 per 1M tokens per hour starting January 1, 2027. Grounding with Google Search is also referenced on the pricing page.
Next step: If you cache large contexts in Gemini, re-estimate that cost line before the January change.
Azure OpenAI documents the Sora video generation model (Sora 2025-05-02)
The Sora (2025-05-02) model creates realistic video scenes from text prompts; Azure provides a Video generation quickstart and concepts guide. Spotlighting for prompt shields (preview) is also noted in the same release notes.
Next step: Optional: follow the quickstart if short AI-generated video fits your content plans.
Amazon Bedrock Agents becomes 'Agents Classic' and closes to new customers
Amazon Bedrock Agents (launched November 2023) is renamed Amazon Bedrock Agents Classic and will no longer be open to new customers; the notice shows June 30, 2026 in connection with its maintenance mode.
Next step: If you run Bedrock Agents, read the Classic maintenance notice and plan your move to the current Agents offering.
Mistral retired OCR 4.0 on September 30, 2026 — move to OCR 4.1
OCR 4.0 (mistral-ocr-4-0) is deprecated and retires September 30, 2026; use OCR 4.1 (mistral-ocr-4-1 or mistral-ocr-latest). Leanstral 1.5 also retires September 30, 2026, and Z.ai GLM 5.2 retires October 31, 2026 (use GLM 5.3, same price).
Next step: Update mistral-ocr-4-0 calls to mistral-ocr-latest now; check GLM 5.2 before October 31.
xAI's grok-voice-transcribe-1.0 reached end of life on October 2, 2026
grok-voice-transcribe-1.0 is deprecated and reached end of life on October 2, 2026; requests to that slug are routed to xAI's current voice models. A September 25 safety_identifier request-field change is also listed.
Next step: Point voice transcription at the current Grok voice model so calls keep working.
Cohere retired legacy model IDs — requests now fail (embed-english-v2.0 among them)
Effective April 4, 2026, the listed Cohere models are no longer available and requests using these model IDs will fail. embed-english-v2.0 is shown among the retired models.
Next step: Search your code for retired Cohere model IDs, starting with embed-english-v2.0, and upgrade.
Microsoft restates Copilot data-residency commitments
Microsoft Copilot is upholding data residency commitments as outlined in the Microsoft Product Terms and Data Protection Addendum, with extensibility documentation linked from the same notice.
Next step: No action unless your data-residency posture is under review.
Azure OpenAI lists GPT-4o-Realtime-Preview data-zone SKUs
GPT-4o-Realtime-Preview-2024-12-17 appears with US/EU Data Zones, Regional, and Global variants, alongside GPT-4o-Mini-Realtime-Preview-2024-12-17 variants on the Azure pricing page.
Next step: If you deploy realtime voice, confirm your region and data-zone choice matches the new SKUs.
Copilot in Excel now links chat answers to the exact workbook changes (Aug 26 – Sep 22, 2026)
In Copilot in Excel, responses now contain links that highlight where changes were made, and navigate directly to new sheets, tables, ranges, and objects like charts and shapes across Windows, Web, and Mac.
Next step: Let your spreadsheet users know; it makes Copilot's edits much easier to audit.
Copilot draws usage from licensed users' allowance; code review for unlicensed users is off by default
Usage from existing licensed users continues to draw from their included monthly allowance as it does today (from June 1, 2026). Copilot code review for users without a Copilot license is off by default; an admin must enable two separate policies to turn it on.
Next step: Admins: leave the code-review policies off unless you have budgeted the extra usage.
Vercel AI Gateway adds Laya (free through Oct 31, 2026) and Microsoft AI models
Laya from Convai Innovations is available on Vercel AI Gateway and free through October 31, 2026 when served by Boundless. Microsoft AI models are also now available on AI Gateway.
Next step: Worth a quick test if you prototype AI features on Vercel.
Subscribers
Demo MRR $187/mo across 3 active companies.
| Company | Vendors | Tier | Status |
|---|---|---|---|
| Harbor & Finch Accounting | 6 | Radar 15 | active |
| Bluepeak Logistics | 4 | Radar 5 | active |
| Marigold Hospitality Group | 9 | Radar 15 | active |
New sign-ups from the onboarding flow appear here as “pending checkout” until their Shopify subscription activates.
Pipeline state
| Source registry | 56 public changelog / pricing / security sources |
|---|---|
| Detection | Last run 10/3/2026 — 45/56 sources ok, 15 new items |
| Brief composer | Draft → owner review → send |
| Payments | Shopify — live (shop.aichangeradar.com) |