Four top labs ship in one week: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark and Claude updates
testingcatalog · x · 2026-09-06
TestingCatalog recaps a rare week with new model releases from four major labs (some unverified):
- OpenAI: GPT-6 Astra rolled out to ChatGPT paid plans and major cloud APIs, with faster computer use, new Codex context memory, and stronger safeguards; OpenAI calls it its first model to cross a critical cybersecurity threshold.
- Google: Released Gemini 3.8 Flash for coding and agents, plus a Flash Cyber variant restricted to vulnerability discovery and automated patching; Gemini desktop is testing Ask/Assign modes and Obsidian integrations.
- Anthropic: Testing Hub Mode in Claude mobile for managing sub-agents, release timing unknown.
- Meta: Muse Video entered closed beta with native audio and strong temporal consistency in 10-second generations; the autonomous Hatch agent is under waitlist with early code for image creation and research tasks.
A secret Muse agent waitlist link ships with the newsletter.
More from Models
- Rumor: Astra Low to X-High Shows Only 1% Reasoning Difference — BLUECOW009 · 2026-09-06
- Unverified "GPT-6 Astra" rumor goes viral: from next-word predictor to frontier model — DeryaTR_ · 2026-09-06
- "Best model ever, yet I have to babysit it": dev bemoans agentic reliability gap — yacineMTB · 2026-09-06
- Cybersecurity pro says Claude guardrails block legit work, asks about Astra — GVTZ59 · 2026-09-06
- Fable 5.1 drops medical refusals, but its 85% life-sciences claim draws scrutiny — JeremyNguyenPhD · 2026-09-06
- ChatGPT Pro's free usage reset also pushes back your weekly limit reset date — dmpizzle · 2026-09-06