GPT-6.1 Sol arrives: dots, Codex plans and what the limits mean

GPT-6.1 Sol is OpenAI's new model for complex work in Codex and ChatGPT Work. The official changelog confirms its September 29, 2026 release. Alongside that launch, current documentation describes dots and a wider set of Pro plans. The sources establish today's model release, but do not establish that every feature and plan change below launched today.

• Model: GPT-6.1 Sol is rolling out across eligible paid plans.
• Dots: continuing work through an agent powered by GPT-6 Astra.
• Access: dots eligibility depends on plan, region and rollout.
• Plans: Pro currently has $100, $200 and $500 monthly options.
• Usage: faster modes consume allowances at different rates from purchased credits.

For the earlier model lineup, see our GPT-6 Sol, Luna and Opus 5.5 comparison. This article covers documentation checked on September 29, not results from hands-on testing.

Where can you use GPT-6.1 Sol?

The launch rollout includes Plus, Pro, Business, Enterprise and Edu in desktop Codex, the CLI, and ChatGPT Work on web and mobile. Free and Go are excluded at launch. Enterprise and Edu administrators must enable the model. Standard and Fast are available; GPT-6.1 Sol Ultrafast is coming later. See model availability.

The model identifier is gpt-6.1-sol. In the CLI, select it with codex -m gpt-6.1-sol when your account has access. OpenAI places Sol in Work and Codex, with no access in Chat.

OpenAI describes its complex-work performance as close to Astra at a lower cost. That is the vendor's positioning. It does not establish that your existing automation needs a new model, or that switching will reduce the total effort needed to accept its output.

What are dots, and who can use them?

Dots are agents powered by GPT-6 Astra with their own cloud computer and browser. They use saved context to carry work between conversations, then return with results or decisions that need your input. Cloud work can continue while your device is off. Access is rolling out gradually under these conditions, according to the dots documentation:

Dot conversations do not consume ChatGPT usage limits. Work or Codex tasks that dots start or manage consume those products' limits normally. Plans also include a deeper-work allowance.

There is an operational detail worth checking before leaving work running: pausing a dot stops its main task. Delegated tasks and recurring schedules need separate controls. OpenAI documents these under stopping dots work. A pause is not a rollback of completed actions.

What do the current ChatGPT Pro plans include?

OpenAI currently lists Pro at $100, $200 or $500 USD per month. Pro has no five-hour limit, but weekly limits may still apply. ChatGPT Work and Codex share usage. These are the current terms in Codex pricing, not a claim that Pro provides unlimited agent work.

The pricing page gives estimated message ranges for some plans. Task complexity and model choice affect actual consumption, so a message count is a poor planning unit for a long project. Check the usage dashboard; the CLI also exposes remaining limits through /status.

My proposed buying rule is to identify the constraint first. A missing feature, a regional restriction and an exhausted usage allowance are different problems. Compare the relevant entitlement before moving to a more expensive plan.

How do Fast and Ultrafast affect usage?

For supported models, speed modes apply different multipliers to included usage and paid consumption. Each figure below is relative to Standard for the same model. The speed documentation distinguishes the two:

These are billing and allowance multipliers. An 8x usage multiplier does not mean a task finishes eight times faster. Astra Ultrafast requires Pro $500 or eligible Enterprise or Edu access; workspace restrictions also apply. It is a separate availability question from GPT-6.1 Sol's launch.

What would I check before switching?

I would compare one completed task with a known acceptance standard on the current setup and GPT-6.1 Sol. Keep the inputs and tools fixed. Record elapsed time, allowance consumed and corrections needed before the output is usable.

For dots, I would start with one ongoing responsibility, a clear update destination and explicit boundaries on actions. Then verify where delegated work appears and how to stop it. Our agent monitoring article explains what a useful execution record should show.

That is a proposed evaluation. GAT has not measured GPT-6.1 Sol performance or dots reliability.