Report: OpenAI may ship next-gen Astra, Sol and Cyber as soon as today
The Information says they could land Friday; a 'gpt-6.2-astra' tag already leaked in usage logs. Also: GPT-Live dials real phones, Copilot adds metrics.

Copy markdown
GPT-6.2 Astra, Sol and Cyber reportedly days away
The Information reports OpenAI could ship next-gen Astra, Sol and a cyber-focused model as soon as today, and a 'gpt-6.2-astra' tag has already leaked on the ChatGPT usage-analytics page and in Codex logs. OpenAI has confirmed nothing and timelines can slip, but if it lands you'll see new frontier options in Codex and the API within days.
Copilot's metrics API now scores your code reviews
GitHub's Copilot usage-metrics API added code-review details and an 'addressed rate' on Nov 5, so teams can finally measure how often Copilot's review comments actually get acted on. A companion update flags telemetry and version gaps, surfacing seats on stale clients that quietly skew your adoption numbers.
GPT-Live can now cold-call a phone number
OpenAI made GPT-Live outbound calling generally available on Nov 4: point it at a Twilio or Telnyx SIP trunk and your agent places real two-way calls, DTMF keypad tones included, for $0.05/min on the voice layer (carrier fees extra). 'Build a voice agent that phones customers' is now a weekend project, not a telephony integration.
Google’s recipe to stop overpaying for agent runs
A new Google developers post shows how to route each task across Gemini models and thinking levels using cheap classifiers, trimming the bill on autonomous coding agents even as aggregate per-token spend keeps climbing. The pattern is model-agnostic: a small router that decides when you actually need the expensive model.