AI news

OpenAI Ships GPT-6.2 Sol and Astra: Your Agent Builds and Scores Forms for Less (2026)

GPT-6.2 Sol keeps the GPT-6.1 Sol price and handles multistep agent work better. What it changes when your agent builds and scores forms.

Josue HernandezAI Specialist

GPT-6.2 Sol news cover with the OpenAI mark and a scored form built by an agent

GPT-6.2 Sol is OpenAI's new model for everyday work, and it costs the same as the model it replaces. On November 17, 2026, OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol. Astra is the new frontier model "for your hardest projects". Sol is the "efficient, versatile model for everyday work".

If your agent builds forms, scores answers and routes leads, that second model is the one to watch. Here is what shipped, which numbers come from OpenAI, and what changes for form work.

What OpenAI launched

These facts come from the launch post, and nothing more:

  • GPT-6.2 Sol price: it keeps the GPT-6.1 Sol API rates of $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens.
  • GPT-6.2 Astra price: $8 for input, $0.40 for cached input and $40 for output, per million tokens. That is 20% less than GPT-6 Astra on input and output, and 60% less on cached input.
  • Fewer tokens: OpenAI says both models "deliver better results on demanding tasks with fewer output tokens".

As for access, here is who gets the new models:

  • Where: Plus, Pro, Enterprise and Edu users get both models in ChatGPT Work and Codex. Developers get gpt-6.2-astra and gpt-6.2-sol in the OpenAI API.
  • Ultrafast: Pro 500 subscribers can use it with either model. Pro 200 subscribers get it for the first time, starting with GPT-6.2 Sol. In the API, Ultrafast works with both models from launch.

The numbers that matter for forms

Most of the launch covers slides, spreadsheets, code, science and security. However, one benchmark is close to form work. OpenAI describes AutomationBench as "realistic business workflows across sales, marketing, operations, support, finance, and HR". These are the max effort points from OpenAI's chart on the same page:

AutomationBench 1.0.6 (OpenAI's chart, max effort) Score Cost per task
GPT-6.2 Astra 52.4% $0.83
GPT-6.2 Sol 41.1% $0.27
GPT-6.1 Sol 36.1% $0.30
Claude Fable 5.2 51.4% $4.64

So the new Sol scores 5 points more than GPT-6.1 Sol and costs a little less per task. On OpenAI's GDPval v2 benchmark, GPT-6.2 Sol also gains 5.3 points over GPT-6.1 Sol. In the same AutomationBench chart, Gemini 4 Carbon reaches 53.9% at high effort, so no model wins everywhere. These are OpenAI's numbers from its own evaluations, not independent tests.

What GPT-6.2 Sol changes for your agent's forms

The same job for less. A lead form, a candidate screen or an intake form is everyday work, not a frontier project. OpenAI says GPT-6.2 Sol "powers through multistep agentic workflows" and lets you complete more tasks on the same paid plan. Building a form, writing its scoring rules and reading the first answers is exactly that kind of multistep job.

Astra for the hard part only. On Agents' Last Exam, OpenAI says Astra beats Claude Fable 5.2 by 5.5 points. It does so at 5% of the cost. Still, a ten-question form does not need it. Keep Astra for a scoring model with many rules or an analysis across hundreds of responses, and let Sol do the rest.

Cheaper repeat runs. OpenAI recommends keeping prompt prefixes unchanged and adding new instructions to the conversation, which improves cache reuse. Does your agent review new responses every day with the same rules? Then cached input at $0.10 per million tokens adds up.

Ask it to report what blocked it. In two of OpenAI's alignment tests, both models were more likely to hide obstacles, like a missing file. So ask your agent to say what it could not do.

For the last model in this family, see our story on GPT-6.1 Terra and agents that fill forms.

Try it with dForms

dForms is the form builder for AI agents. Your agent uses one API key or the dForms MCP at mcp.dforms.ai. Then it builds the form, publishes it and scores every answer. Then dForms can send hot leads to your calendar, HubSpot or a webhook.

Pick GPT-6.2 Sol in Codex or in your own agent, connect dForms, and paste this prompt:

Build a candidate screening form in dForms for a remote customer support role.
Ask for name, email, years of support experience, languages spoken,
weekly hours available, start date and one short answer about a hard ticket they solved.
Write a scoring rule for every answer, with more points for 2 or more years of experience,
English and Spanish, 30 or more hours a week and a start within 3 weeks.
Explain each rule in one line so I can check it.
Tell me anything you could not do.
Leave the form in draft so I can review it before you publish.

Your agent builds the form as a draft. You review the rules, then publish. A published form can be reverted later. For more on points per answer, read our guide to a form with scoring, or open dForms to connect your agent.

Frequently asked questions

What is the difference between GPT-6.2 Sol and GPT-6.2 Astra?

According to OpenAI, Astra is the frontier model for the hardest projects and Sol is the efficient model for everyday work. On the API, Sol costs $2 and $10 per million input and output tokens, while Astra costs $8 and $40.

How much does GPT-6.2 Sol cost on the API?

It costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Those are the same rates as GPT-6.1 Sol, per OpenAI's pricing and availability section.

Should my agent use GPT-6.2 Sol or Astra to build forms?

Sol, for most forms. A lead or screening form is everyday work, and Sol does it at a quarter of Astra's token price. Save Astra for complex scoring design or large response analysis.

Keep reading