GPT-6.2 Astra vs. Sol: Which model should you use for each task?

Choose GPT-6.2 Astra when evidence conflicts or the cause of a bug may span multiple services. GPT-6.2 Sol is a good starting point for drafting and code review when the deliverable and scope are clear. OpenAI announced both models on November 17, 2026.

GPT-6.2 Astra and Sol at a glance

Comparison GPT-6.2 Astra GPT-6.2 Sol
Typical tasks Reconcile conflicting evidence or diagnose a cross-service failure Draft text or review and revise a bounded change
When to choose it Several explanations or constraints need to be weighed together The goal and acceptance criteria can be defined up front
Standard API rates per million tokens Input $8 · cached input $0.40 · output $40 Input $2 · cached input $0.10 · output $10

What work is GPT-6.2 Astra suited to?

If two reports reach different conclusions about the same policy, ask Astra to align their study periods, samples and measures in a table. Have it point to the passage that supports each conclusion and flag differences the reports cannot explain. That gives you a list of questions to investigate.

For a missing notification after a successful payment, provide reproduction steps, sanitized logs and relevant tests. Ask it to trace where the event was created, forwarded and received, then verify both the normal payment flow and a retry after the fix.

When is GPT-6.2 Sol enough?

To turn meeting notes into a plan, ask Sol to extract decisions, owners, deadlines and open questions. Tell it to leave unspecified owners and dates as undecided so you can fill them in during review.

For code review, supply the changed files and expected behavior. Ask for affected tests and comments tied to specific changes. If the issue leads to another service or an unresolved requirement, Astra becomes a reasonable choice for the wider investigation.

If you want to tune reasoning within a single model, see the GPT-6.1 Sol High and Medium comparison for ways to compare review depth, wait time, and usage.

How do performance and cost compare?

In Cognition’s FrontierCode 2.0 results reported by OpenAI, Sol scored within 0.5 points of Astra at about one-fifth the cost when both used Max reasoning. The evaluation measures whether changes to real codebases meet maintainers’ standards for merging. The figures apply to that evaluation setup. (OpenAI announcement and API rates)

Astra’s standard API token rates are four times Sol’s across all three categories in the table. What you pay per task depends on input, cached input and output volume, plus retries. For a recurring workflow, compare both models on the same inputs and acceptance criteria and track the total cost and revisions needed to approve the result.

Where can you use them in ChatGPT and Codex?

OpenAI’s launch guide places both models in ChatGPT Work and lists the general Chat models separately. Both are available in the Codex app and CLI. Access depends on your plan and, where applicable, workspace settings.

The Work launch guide lists Plus, Pro, Business, Enterprise and Edu. Enterprise and Edu admins must enable the models in workspace settings first. (ChatGPT Work and Codex launch guide)

The API model IDs are gpt-6.2-astra and gpt-6.2-sol. The table shows standard API rates. For usage in Work or Codex, check the terms of your subscription. (OpenAI API model comparison)