GPT-6 Astra ships. Gemini Enterprise Workflow Builder goes GA.

OpenAI shipped GPT-6 Astra (gpt-6-astra) on September 3 — after Issue 9’s morning close, which only had Path to Astra “available soon” with no API id. 1.05M context, $10/$50 Standard, Critical cyber under the Preparedness Framework; all benches vendor-flagged. Trusted Access enterprises first; Plus/Pro/Business/Enterprise + API + Azure + Bedrock over coming days; Enterprise off by default. Same window: Gemini Enterprise Workflow Builder (formerly Agent Designer) GA, plus Latency/Error Observability views; Databricks Genie One web search (Beta). Local: NVIDIA Personal AI Router (PAIR) beta at IFA — Mac floor M4+ (M3 Ultra out) — and Ollama v0.33.3 brings gemma4 images/audio to the MLX engine (Issue 9 catch-up). No new OSS tools or models GA. From the field empty this issue.

Window after Issue 9 close (~2026-09-03 ~08:15 AM ET) through 2026-09-04 morning. Issue 9 already covered NVIDIA↔HF $12.93B, Claude Fable 5.1, Gemini 3.8 Flash, llama.cpp b10785. Primary sources only. Vendor evals flagged.

Breaking AI News

Frontier

1 story

GPT-6 Astra (gpt-6-astra)

1.05M ctx, 128k out, $10/$50, Critical cyber, Responses-required tools. Vendor evals flagged. vendor

1 story Open section

Enterprise

3 stories
3 stories Open section

Local

2 stories
2 stories Open section

MLX

1 story
1 story Open section

Try this on a Mac

1 story
1 story Open section