MODELS

GPT-6.2 Sol

OpenAI’s upgrade to GPT-6.1 Sol, positioned below the flagship GPT-6.2 Astra — an efficient, versatile model for multistep agentic workflows and clear, natural writing.

Released
17 Nov 2026
Developer
OpenAI
Context window
1.05M tokens
Modalities
  • Text
  • Image
Pricing / 1M
$2 in · $10 out
Reasoning depth
  • Low
  • Medium
  • High
  • xHigh
PROMPT
341 WORDS

The Right to Overrule the Machine

A billion people with access to advanced AI is not the same thing as a billion people with more power. A teacher who can translate a letter for a parent has gained power. A teacher who must spend her evenings correcting lesson plans generated by a system the district bought has not. The difference is whether she gets to decide what the machine is for.

That should be the test of the Mass Intelligence era. I want AI to make expertise easier to reach and authority harder to hide. If a hospital uses AI to deny care, it should have to explain the decision and give the patient a timely appeal to someone who can reverse it. If a company saves a day of work every week with AI, workers should have a say in whether that dividend comes back as time, pay, or a higher quota. Otherwise, we’ll call it progress because we’re measuring what the machine produces, not what people get.

Schools should stop treating polished output as proof of learning. Let students use AI, then ask them to defend a claim aloud, show where the model misled them, and say what evidence would change their minds. Keep some work deliberately unaided. You can’t meaningfully delegate a skill you’ve never practiced.

And we should quit assuming that because an AI is useful in a conversation, it deserves the keys to a system. The more irreversible an action, the less autonomy we should grant. Helping someone understand a medical result is one thing; changing their treatment without anyone accountable signing off is another. Put that distinction into law and software, not just ethics statements.

The future I want is one where a tenant can read a lease without hiring a lawyer, a nurse gets an hour back with patients, and either can challenge a decision made about them. The real test of Mass Intelligence is not how often the machine is right. It’s how much room people have to act when it’s wrong—and how much they gain when it’s right.