GPT-6.2 Explained: Astra, Sol and Who Gets Access

Find GPT-6.2 Astra and Sol in ChatGPT Work, Codex and the API. See who gets Ultrafast and which jobs justify switching models.

Published

GPT-6.2 Explained: Astra, Sol and Who Gets Access

Start with GPT-6.2 Sol for everyday coding and writing. Reach for GPT-6.2 Astra when the hard part is figuring out what happened across several services, documents or tools. OpenAI released both models on November 17, 2026, in ChatGPT Work, Codex and the API.

What changed in GPT 6.2?

GPT-6.2 Astra improves on GPT-6 Astra at drawing a conclusion from a set of documents and presenting it in the format you asked for. Consider a founder deciding whether to roll out a payments feature. In OpenAI’s fictional Shared Bills experiment, the older Astra led its memo with an increase in transactions using the feature. The newer Astra led with the finding that mattered for rollout: the experiment showed no clear increase in overall card use or payments between people. For the founder, that distinction changes whether the feature has earned a wider rollout. OpenAI dates both models to the same entry in its November release notes.

OpenAI also reports that Astra follows document and slide templates more closely, writes cleaner code and handles finer details in computer-aided design tools. On Agents’ Last Exam, which tests professional assignments that take many steps, it scored 5.2 points above GPT-6 Astra.

GPT-6.2 Sol is the everyday upgrade from GPT-6.1 Sol. OpenAI describes clearer writing and better performance on jobs that take several steps, with fewer output tokens, the chunks of text the model generates. That efficiency lets you finish more work within the same paid plan. Sol came within 0.5 points of the new Astra on FrontierCode 2.0, a test of whether code changes meet maintainers’ standards for merging. Both models used maximum reasoning effort for that comparison. That is a good reason to try Sol on your review queue even if you use Astra for difficult bugs.

For earlier versions, the GPT-6 family guide covers the wider lineup. If you already split coding work between models, the previous Sol upgrade and the Sol and Luna comparison give you the baselines for that decision.

Where can you use Astra and Sol?

In ChatGPT, open Work. Both models are available on Plus, Pro, Business, Enterprise and Edu in Work on web, desktop and mobile, and in the Codex app and command-line interface (CLI). OpenAI’s plan guide explicitly includes Business. The access details here were checked on November 18, 2026. Plan and client availability

The selection controls follow the product you open.

Where you workHow to select GPT-6.2What determines access
ChatGPT WorkIf you see the Power slider, open Advanced to pick Astra or Sol and choose reasoning effort.Your plan, access to Work and workspace settings.
Codex app or CLIUse the app’s model picker or type /model in the CLI.With ChatGPT sign-in, check your seat and Codex access. With API-key sign-in, check the key’s organization and project.
Responses APISet the model ID to gpt-6.2-astra or gpt-6.2-sol.The API project making the request.

New Work chats on supported consumer plans start with Sol at Light reasoning effort. In Enterprise workspaces with access enabled, the default is Sol at Medium. Work and Codex share usage, so moving a task from one to the other still draws on the same allowance. GPT-6.2 product guide

Enterprise and Edu admins: Both models start switched off. Open Admin Console → Workspace settings → Models and enable the models you want members to use. Confirm that the member can open Work or Codex as well. A missing model in someone’s picker is a reason to check their seat, workspace and client before changing their plan. Workspace model settings

Who gets Ultrafast?

Ultrafast makes the selected model generate its answer faster. Choose it when someone is waiting on the output, such as a developer iterating on a patch. Reasoning effort is a separate choice. In Work and Codex, Pro 200 gets Ultrafast with Sol. Pro 500 gets it with either model. The faster setting uses more of your allowance.

Eligible Enterprise and Edu workspaces can use Ultrafast with either model, but an owner must separately turn it on in workspace permissions for selected users or the whole workspace. Enterprise needs a credit-based or USD usage-based agreement. Edu needs a credit-based plan. Ultrafast eligibility and controls

An open model-access gate leads to a standard trail and a separately locked Ultrafast entrance
A workspace owner can enable a model while keeping Ultrafast switched off.

API developers get Ultrafast with both model IDs at launch. If your project requires EU data residency, use Standard, Fast or Batch. GPT-6.2 Ultrafast supports US residency only. For budget decisions, see the OpenAI API pricing guide and ChatGPT plan guide.

Should you switch?

Agent builds. Start with Sol for a repeatable sequence, such as collecting project updates and drafting a weekly digest. Try Astra when the agent has to resolve conflicting inputs before it can act. Give it a completed case where the ticket and the underlying record disagree. Check whether it catches the discrepancy before recommending a change.

Coding. Give Sol the pull request when a reviewer has already named the defect and the expected behavior. Give Astra the failing workflow when you still need to trace the cause across services. Judge the patch by whether the affected tests pass and the surrounding workflow still works.

A walking trail leads to Sol for a known fix, while an Astra climber inspects a chain of connected anchors
A known correction is a Sol task. Tracing a failure across components is a reason to try Astra.

Long documents. Try Astra for a board memo assembled from several reports and a house template. Ask it to trace the headline finding back to the relevant source and preserve any qualification that could change the decision. For a copyedit with the conclusion already agreed, choose Sol.

Computer use. Try Astra when success depends on small visual details, such as checking how components fit in a mechanical drawing. Give it a copy of the design and a specific joint to inspect. Have the engineer verify that joint before accepting the revision.

Everyday writing. Use Sol for meeting notes, project briefs and messages to customers. A good first assignment is turning a discussion into decisions, owners and open questions. If your current workflow already produces a sendable draft, keep it until a side-by-side read shows less editing.

What the cyber scores leave out

Both models meet OpenAI’s Critical cybersecurity capability threshold. OpenAI reports improvements in secure code and patches, and says its internal cyber benchmark runs disabled safeguards and classifiers. The released models have tighter safeguards for higher-risk assistance. Those restrictions can affect legitimate vulnerability research and authorized security testing.

If you lead a security team, evaluate routine defensive work separately from research that can encounter those restrictions. A useful first assignment is reviewing a proposed patch for a known issue and checking that the change preserves intended behavior. A benchmark score alone cannot tell you which requests your staff can complete in the released product.

Get the newsletter for practical guidance on model changes that affect your work.

Published
Category
AI
Related Articles
Kimi vs Qwen

Kimi vs Qwen

Kimi K3.1 Pro vs Qwen4-Max for coding agents and products: current access, long context, open weights, API tradeoffs and US deployment choices.Nov 17, 2026AI
Qwen 4: Qwen4-Max Explained

Qwen 4: Qwen4-Max Explained

Qwen4-Max explained: production versus preview access, context and output limits, open-model promises, and when to move from Qwen3.8-Max.Nov 17, 2026AI
Perplexity Alternatives 2026: ChatGPT, Gemini, Claude, Consensus, Kagi, You.com and Perplexica

Perplexity Alternatives 2026: ChatGPT, Gemini, Claude, Consensus, Kagi, You.com and Perplexica

Compare ChatGPT, Gemini, Claude, Consensus, Kagi, You.com and Perplexica by research job, source access and practical limits.Nov 16, 2026AI
Plaud Alternatives in 2026: Otter, Granola, Fireflies, Krisp, MeetGeek, TicNote and Omi (Compared)

Plaud Alternatives in 2026: Otter, Granola, Fireflies, Krisp, MeetGeek, TicNote and Omi (Compared)

Compare Plaud alternatives for in-person meetings and online calls: Otter, Granola, Fireflies, Krisp, MeetGeek, TicNote and Omi.Nov 16, 2026AI
Gemini 4 Hydrogen Explained: Everyday Coding and Agent Work

Gemini 4 Hydrogen Explained: Everyday Coding and Agent Work

Where Gemini 4 Hydrogen fits beside Carbon and Argon, how to access it, and which thinking, context and migration settings matter for everyday work.Nov 14, 2026AI
AI Tools for HR in 2026: Workday, Rippling, BambooHR, HiBob, Leapsome, Lattice, Checkr and Trainual (Compared)

AI Tools for HR in 2026: Workday, Rippling, BambooHR, HiBob, Leapsome, Lattice, Checkr and Trainual (Compared)

Compare eight AI tools by HR job, with verified features, plan requirements and human checks for hiring, onboarding, reviews and people analytics.Nov 13, 2026AI
Fish Audio vs ElevenLabs

Fish Audio vs ElevenLabs

Choose Fish Audio or ElevenLabs for voiceovers, cloning, an app or a phone agent. Current models, commercial rights and migration limits compared.Nov 13, 2026AI
Claude Sheets

Claude Sheets

Claude Sheets builds shared trackers, budgets and plans. See beta access, outlined edits, export options and when you still need Excel or Google Sheets.Nov 13, 2026AI
Newsletter

One letter, every Sunday.Working systems, not hot takes.

Weekly. No spam. Unsubscribe anytime.