Skip to content

Vibe Check: GPT-6.2 Astra and Sol Leave Us Wanting More

GPT-6.2 Astra and Sol handle bounded tasks efficiently. The work we most want help with still demands more judgment and follow-through.

November 17, 2026

At the start of this year, Mike Taylor, Every’s head of evals, was testing AI models on a set of tasks that none of them could do well: a training curriculum built around a theme, a minute-by-minute plan for running a show, and exercises for executives, each with its own prompt and realistic data. OpenAI’s two new models, GPT-6.2 Astra and GPT-6.2 Sol, handle all of them. Their predecessors already did well on these tasks, too.

“These tasks were all basically unsolved at the beginning of the year... now Codex has kinda saturated them,” Mike says. “Which is great to see. But leaves me wanting more.”

Subscribers only

Only available for paid subscribers

Subscribe to read the rest of this article.

Start free trial →

Already have an account? Log in.

Uploaded image


A virtual summit for the AI era of finance

Finance teams are tokenmaxxing, but for all the right reasons. Product managers, engineers, and builders at large are emerging across AR, AP, payroll, FP&A, and other core finance functions, with an ardent appetite for reimagining every tedious workflow with AI.

Traceable by Chargebee is a virtual summit spotlighting that transformative work.

Tomorrow, join CFOs, controllers, and senior finance operators as they share hands-on notes on pairing human judgment with the apps, real-time dashboards, and complex agentic workflows they've been shipping. The summit features leaders from Vercel, PostHog, Public.com, Every, Campfire, and other frontier teams.

Katie Parrott is a staff writer. She writes Working Overtime and contributes to Vibe Checks, Source Code, and Context Window.

We use analytics and advertising tools by default. You can update this anytime.