π AI Watch

DAILY DIGEST

October 28, 2026

7 stories · 9 in brief · 4 min read

Try

Useful tools & techniques

Copilot adds image generation in three developer surfaces

GitHub Copilot’s public preview adds natural-language image generation and editing, powered by GPT Image 2.5 Flare. It’s available to Pro, Pro+, Business and Enterprise users in VS Code, the Copilot app and Copilot CLI; previews appear inline and generated images are saved to the workspace.

Try it for quick visual assets or edits inside an existing Copilot workflow.

GitHub open-sources design-system prototyping tool Replica

Replica is an MIT-licensed AI prototyping tool in the GitHub Copilot app. Natural-language prompts turn a team’s components, patterns and documentation into interactive prototypes; an optional self-hosted Replica Hub supports sharing and comments.

Test it against your design system when exploring UI ideas before implementation.

Nvidia releases open-weight speech transcription model

Nvidia released free, open-weight Nemotron 3.5 Transcribe, a 30B-parameter model that transcribes audio in up to 40 languages and translates 31 languages to and from English. It supports recordings up to an hour and separates up to eight speakers. Nvidia also released version 2 of its open Granary speech dataset.

Evaluate it for multilingual batch transcription or speaker-separated audio workflows.

ServiceNow actions arrive in OpenAI’s marketplace

A ServiceNow MCP plugin is available in the OpenAI marketplace, allowing ChatGPT to take governed ServiceNow actions under existing permissions. The companies also announced planned Codex, cybersecurity and voice integrations; those are not described as available yet.

Check the plugin if your team uses ServiceNow and needs permission-governed actions from ChatGPT.

Know

What changed

Google previews connected apps for Gemini and Search

Google’s Connected Apps Platform is in public preview for US developers through a waitlist. Businesses can submit apps for Gemini, Search or both using a cloud-hosted MCP server or Android AppFunctions, registered through Google Connected Console. Separately, CEO Sundar Pichai said Google plans to bring enterprise Gemini agent features to Search and the Gemini app, then Chrome, Android and Workspace. He reported nearly 32 billion first-party model API tokens processed per minute.

If you build an app integration, check the preview requirements and waitlist before planning a launch.

Artificial Analysis updates model and coding benchmarks

Artificial Analysis Intelligence Index v5 now uses private test sets or held-out answers for 60% of its evaluations. It adds a repository-maintainer-judged coding benchmark and agentic science tasks, and updates several existing tests. The evaluator ranks Claude Opus 5.5 first overall, ahead of GPT-6 Astra; its new Progress Index tracks capability over time.

Use the revised results as one comparison point, and check which task fits your workload.

Watch

Signals to keep an eye on

Gemini 4 Carbon posts strong coding scores, with limited access

Google announced Carbon two weeks after Argon, calling it a stronger model with advanced cybersecurity capabilities; access is currently limited to selected cyber defenders and trusted partners. Artificial Analysis ranks it first on its Intelligence Index v5 and second on its Coding Agent Index, tested with Antigravity CLI. The evaluator reports a 1M-token context window and standard pricing of $4/$20 per million input/output tokens, with a 50% launch discount and no confirmed end date.

Treat the coding result as a promising signal, not a deployable option; access remains restricted.

In brief

Also noted today

All sources

Everything captured for this issue

Articles 1

X posts 3

Links 29

Updated 2026-10-29 13:06 UTC

Get AI Watch by email

A compact digest when Nick publishes. No extra noise.

Confirm by email. Unsubscribe any time.