OpenAI began rolling out GPT-6 in ChatGPT on 7 October, first to Plus, Pro, Business and Enterprise subscribers in the Chat tab, with Free and Go users following on 8 October.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source Paid tiers get GPT-6 Sol; free tiers get GPT-6 Luna.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source Both arrive inside a new response format OpenAI calls Intelligent UI, which places tappable buttons, calculators, interactive charts and editable graphs directly inside answers when the model judges a visual will help.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source The models behind Work and Codex are unchanged.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source

The launch is notable for what OpenAI chose to emphasise. The only performance figure attached to it is a 44 percent improvement in time to first token, a measure of how quickly an answer starts rendering rather than how good it is.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source This is a product launch about the interface, not a benchmark launch about the model.

Why the interface is the bet

An OpenAI product manager told TechCrunch that ChatGPT has mostly been a text interface, and the demonstrations shown were explanatory: wing lift with diagrams drawn on the fly, bicycle mechanics, hiking maps, a savings calculator.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source Those are not tasks where a smarter model wins by much. They are tasks where a more useful presentation does.

We assess with moderate confidence that this reflects a judgment inside OpenAI that model quality at the consumer tier has reached the point where further gains are hard for users to perceive, and that differentiation now comes from what the answer does rather than what it says. A rendered calculator that a user can adjust keeps them inside ChatGPT for the next step; a paragraph of text sends them elsewhere to act on it.

There is a cost visible in the details. Astra extended thinking mode disables Intelligent UI, even on Pro.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source The company's most capable reasoning path and its new interface do not yet run together, which suggests the rendering layer depends on the model starting to output before it has finished reasoning.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source Users who want the deepest answer and the richest presentation must currently choose.

The quieter launch the day before

On 6 October OpenAI opened a public beta of a Decisions API that runs only on GPT-6 Luna.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source It returns typed answers rather than prose: a probability that a statement is true, a choice among fixed options with confidence, or a score against ordered levels.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source It costs 0.10 dollars per million input tokens, and output is free.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source OpenAI says it is about ten times faster than the Responses API, but has published no benchmark.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source

Put the two launches together and the tiering logic is clear. Luna is the volume model: free consumers in ChatGPT, and developers who need classification and routing at near zero marginal cost. Sol is the paid experience. OpenAI is using one model generation to address two very different price points, and pricing output at zero on the developer side is a direct play for the classification and moderation workloads that currently run on small open models.

What the tier split signals

Giving free users a named GPT-6 model a day after paying users is a choice about distribution. Earlier generations reached the free tier late or in reduced form. Here the gap is one day, and the difference between tiers is expressed as two models rather than as a usage cap on one.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source That makes the free product a showcase rather than a trial, and it puts the burden of the upgrade case on Sol being visibly better at the tasks paying users care about. Enterprise access, meanwhile, depends on administrator settings, so the business rollout will be uneven in practice even though it began on day one.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source

Who gains and who loses

Gains: OpenAI's free tier, which receives a current generation model and the new interface within a day of paid users.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source Developers who run high volume yes or no and scoring workloads, if the Decisions API's accuracy holds up.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source

Loses: providers of small classification models and moderation APIs, now facing free output from a frontier lab. Third party apps whose value was a calculator or chart built on top of a chatbot answer, since the chatbot now builds it.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source

The counter case

The skeptical reading is that a launch led by an interface and a latency number is a launch without a capability story. If GPT-6 Sol were a large step on reasoning, the release would lead with it. On that view Intelligent UI is packaging, and packaging is easy to copy. The interactive elements described are well within reach of any competitor with a rendering layer, and the trade off with Astra suggests the integration is not yet deep.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source The Decisions API, for its part, is in beta with no published accuracy comparison and is missing from Luna's own endpoint support table.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source

Assessment

We assess with moderate confidence that Intelligent UI raises engagement on explanatory and planning queries, and that competitors ship comparable inline rendering within two quarters. We assess with moderate confidence that the Decisions API reaches general availability with output pricing at or near zero, because the strategic purpose of the free output is to displace small model workloads and charging later would undercut it.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source

What to watch

First, whether OpenAI publishes capability benchmarks for Sol and Luna separately before the end of October. Their absence at launch is itself information.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source

Second, whether Astra extended thinking gains Intelligent UI support in a later update, which would show whether the conflict is architectural or temporary.1 AI News Weekly 2026-10-07 Sol for paid tiers on 7 Oct, Luna for Free and Go on 8 Oct; Intelligent UI renders interactive elements in replies; 44 percent faster time to first token; Astra extended thinking disables Intelligent UI. Open source

Third, general availability pricing for the Decisions API, which OpenAI has said will come in the coming weeks.3 AI News Weekly 2026-10-07 Public beta 6 Oct on gpt-6-luna only; predicate, choice or score outputs; 0.10 dollars per million input tokens, output free; claimed 10x faster than Responses API, no benchmark published. Open source

Fourth, whether GPT-6 reaches the Work and Codex surfaces by the end of the year, completing the migration the Chat launch began.2 Let’s Data Science 2026-10-07 Rollout from 7 Oct in Chat; Sol for paid, Luna for Free and Go from 8 Oct; enterprise subject to admin settings; Work and Codex models unchanged. Open source