AI Pulse

What's moving in AI

Official lab announcements, latest model releases from OpenRouter, and the day's most important AI news — curated and updated daily.

Last updated: Jul 9, 2026 · Back to home

From the labs

Announcements

Official posts from AI labs and key researchers, embedded directly.

OpenAI announced GPT-Live is fully rolled out to ChatGPT Go, Plus, and Pro users, with free rollout in progress.

Loading tweet...

OpenAI Developers announced GPT-Realtime-2.1-mini availability in the API with reasoning and tool-use support.

Loading tweet...

OpenAI announced that GPT-5.6 Sol, Terra, and Luna will launch publicly this Thursday with broader preview access.

Loading tweet...

Meta promoted Muse Image as a generative media feature rolling into Meta AI, Instagram, and WhatsApp workflows.

Loading tweet...

Claude announced Claude Cowork for mobile and web, positioning Claude as a persistent cross-device work assistant.

Loading tweet...

Google DeepMind highlighted the Predicting the Past Skill in Google Antigravity, grounding Gemini in Aeneas and Ithaca for historical text work.

Loading tweet...

Latest releases

Latest models

Refreshed directly from OpenRouter's live model catalog. Sorted newest-first by catalog release timestamp; no stale Arena snapshots or copied-forward model lists.

#Model
1
2
Hy3New
3
4
5
6
7
8
9
10
11
12

Today's signal

Latest news

Curated from today's digest — top stories with editorial context.

NewsThe Verge

GPT-Live makes voice the product surface, not just a feature

OpenAI says GPT-Live is now broadly rolled out to paid ChatGPT users, with free rollout in progress. The shift matters because natural turn-taking can move voice AI from demo to daily habit.

Read
NewsOpenAI

Coding-agent benchmarks are hitting their leaderboard hangover

OpenAI audited SWE-Bench Pro, found 30% of tasks broken, and said it no longer reliably measures frontier coding capability. The signal is a pivot from trophy scores toward real-repo, long-horizon evaluation.

Read
NewsTechCrunch

AI provenance becomes a political and media infrastructure problem

Google's deepfake detector system was used to debunk an AI-generated hoax image of Senator Mitch McConnell. Synthetic media detection is becoming fast-moving trust infrastructure for journalists, platforms, and readers.

Read
NewsMistral AI

Embodied and agentic AI gets more specific but still needs proof

Mistral announced Robostral Navigate for robot navigation via natural-language instructions and a single RGB camera. The broader agent narrative is moving into robotics and workflows, but needs measurable autonomy and safety evidence.

Read
NewsArs Technica

AI use in education is becoming a measurement crisis

A Brown University professor ordered an in-person final after suspecting AI cheating, and scores fell 50%. The story highlights the need to separate assisted output from demonstrated competence.

Read
NewsOpenAI

OpenAI queues up GPT-5.6 Sol, Terra, and Luna

OpenAI says GPT-5.6 Sol, Terra, and Luna will launch publicly this Thursday, with preview access expanding globally. The bundle points to workflow/product tiers rather than a single isolated model drop.

Read

Read deeper analysis on the blog →