OpenAI Launches GPT-6.2 Astra and Sol, But Regular ChatGPT Users Can’t Select Them

OpenAI Launches GPT-6.2 Astra and Sol, But Regular ChatGPT Users Can't Select Them
Share

OpenAI has released GPT-6.2 Astra and Sol with improved coding, professional work and reasoning capabilities. The models are available in ChatGPT Work, Codex and the API, but regular ChatGPT conversations are not receiving the upgrade.

OpenAI has introduced two new artificial intelligence models, GPT-6.2 Astra and GPT-6.2 Sol, expanding its latest model family with improvements aimed at professional tasks, software development and complex projects. However, the launch comes with an important restriction: the models are not available in the regular ChatGPT conversation experience.

The company announced the models on November 17, 2026, making them available through ChatGPT Work and Codex for eligible Plus, Pro, Enterprise and Edu subscribers. Developers can also access both models through OpenAI’s API.

The launch introduces a more capable flagship model at a lower price than its predecessor, alongside a cheaper model intended for everyday professional work. GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens through the standard API, while GPT-6.2 Sol costs $2 and $10, respectively.

According to OpenAI’s official GPT-6.2 announcement, the models improve performance across coding, professional tasks and other demanding assignments. But the company’s decision to keep them outside regular ChatGPT conversations creates a clear difference between the models available for chatting and those available for completing longer projects.

GPT-6.2 Astra and Sol Are Available, but Not in Regular ChatGPT

The biggest surprise in OpenAI’s latest release is where the new models can be used.

GPT-6.2 Astra and Sol are available in ChatGPT Work, the company’s environment for handling longer assignments, and in Codex, its coding-focused product. They are also available to developers building applications through the API.

However, users who open an ordinary ChatGPT conversation will not find GPT-6.2 Astra or GPT-6.2 Sol in the standard model selector.

This applies even to subscribers who pay for ChatGPT Plus or Pro. Their subscriptions may provide access to the new models through Work and Codex, but they do not automatically make the models available in regular Chat.

OpenAI confirmed the distinction in its ChatGPT Work and Codex documentation, which explains that the new models can be selected in Work or Codex while ordinary Chat retains its existing choices.

The restriction does not mean all regular ChatGPT users are excluded from GPT-6.2. Eligible subscribers can access the models by switching to the supported experience, provided the rollout has reached their account.

The important difference is between ChatGPT’s products rather than simply between paying and nonpaying customers.

ALSO READ: OpenAI Brings GPT-6 and Intelligent UI to All ChatGPT Users

What Is GPT-6.2 Astra?

GPT-6.2 Astra is OpenAI’s new flagship model for difficult professional and technical work.

The company positions it as a model for assignments that require advanced reasoning, coding, computer interaction and the ability to work across documents and applications.

For example, a developer could use Astra to investigate a software problem across multiple services, identify possible causes and prepare changes for review.

A business user could ask the model to analyze documents, organize information and produce a presentation or website based on the findings.

OpenAI says the model performs better than GPT-6 Astra on several evaluations while using fewer tokens to complete tasks.

This efficiency improvement matters because the cost of running advanced models depends on how much information they process and generate. A model that completes a task using fewer tokens may cost less overall, even if its per-token price remains relatively high.

However, benchmark improvements do not guarantee better results for every user or every assignment. Performance can vary depending on the complexity of the task, the instructions provided and the tools available.

What Is GPT-6.2 Sol?

What Is GPT-6.2 Sol?

GPT-6.2 Sol is the lower-cost model in the new release, intended for users who need capable AI assistance without paying flagship-model prices.

OpenAI describes Sol as suitable for multistep work, coding, writing and tasks that involve several tools or applications.

The model is intended to balance performance and efficiency, making it a practical option for repeated assignments that do not require the highest available level of reasoning.

For example, a user could ask Sol to research a topic, summarize documents, draft content and organize the results into a finished deliverable.

A developer might use the model for routine code changes, testing or investigating straightforward software issues.

OpenAI says Sol approaches Astra’s performance on some coding evaluations while completing tasks at substantially lower cost.

The company also says Sol produces more natural written responses and works effectively across technical and nontechnical assignments.

These claims are based on OpenAI’s reported evaluations and product testing. Independent results may differ depending on the tasks used to compare the models.

GPT-6.2 Astra Is 20% Cheaper Than GPT-6 Astra

One of the most important changes is the reduction in API pricing for OpenAI’s flagship model.

GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens at standard API rates. Its predecessor, GPT-6 Astra, cost $10 per million input tokens and $50 per million output tokens.

That represents a 20% reduction in both standard input and output pricing.

OpenAI has also reduced the price of cached input tokens. GPT-6.2 Astra costs $0.40 per million cached input tokens, representing a 60% reduction compared with the previous model’s $1 rate.

Cached input pricing matters for applications that repeatedly send similar information to a model, such as coding assistants working with the same project files or agents that maintain a large set of instructions.

GPT-6.2 Sol costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. These rates remain unchanged from GPT-6.1 Sol.

At standard API rates, Sol’s input and output tokens cost 75% less than Astra’s.

For example, processing one million input tokens and generating one million output tokens would cost $48 with GPT-6.2 Astra and $12 with GPT-6.2 Sol, before any applicable discounts or additional charges.

Actual costs depend on the amount of information processed, the output generated, caching and the model’s efficiency on a particular assignment.

ALSO READ: OpenAI’s GPT-6 Sol and Luna Launch With 50% Lower API Prices

OpenAI Says GPT-6.2 Can Complete Tasks at Lower Overall Cost

The lower token prices are only part of OpenAI’s efficiency claims.

The company says GPT-6.2 Astra can complete certain professional assignments at less than one-third of the cost per task of GPT-6 Astra.

That improvement reflects a combination of lower API rates, fewer generated tokens and better performance on some complex assignments.

A model that solves a problem with fewer attempts may reduce the total cost of completing the work. This is particularly important for coding agents and research systems that may perform many steps before producing a finished result.

OpenAI also reports improvements on professional and scientific evaluations, including Agents’ Last Exam and coding benchmarks.

The company says GPT-6.2 Astra performs better than GPT-6 Astra on these evaluations while requiring less spending to achieve a result.

However, cost-per-task comparisons depend on the evaluation method. A model that performs well on a benchmark may still require additional attempts or human corrections when used on real projects.

Users should therefore compare the cost of successfully completed work rather than relying entirely on published token prices or benchmark scores.

Why OpenAI Is Keeping GPT-6.2 Out of Regular ChatGPT

OpenAI has not publicly established that GPT-6.2 will never become available in regular ChatGPT conversations.

For now, the company is introducing the models through Work and Codex, where their capabilities can be used for longer assignments involving documents, software and connected tools.

Regular ChatGPT is primarily designed for conversational assistance, everyday questions and shorter interactions.

ChatGPT Work serves a different purpose. It allows users to hand off substantial assignments that may involve research, file analysis, website interaction and the creation of finished deliverables.

Codex focuses on software development, including code editing, debugging and related engineering work.

These environments are better suited to showcasing models that can carry out multiple steps and interact with tools.

However, this product distinction should not be confused with a technical limitation. OpenAI has not said that GPT-6.2 Astra or Sol is incapable of handling ordinary conversations.

The current restriction reflects the company’s launch and product-access decisions.

Which ChatGPT Plans Get GPT-6.2 Astra and Sol?

OpenAI says GPT-6.2 Astra and Sol are available to eligible ChatGPT Plus, Pro, Enterprise and Edu subscribers through Work and Codex.

Plus users can access the models in supported environments, subject to their plan’s usage allowances and the progress of the rollout.

Pro subscribers also receive access, with additional performance options available on certain higher-priced Pro tiers.

Enterprise and Edu customers can use the models when their workspace administrators enable the necessary permissions.

Free and Go users are not included in the announced GPT-6.2 Work and Codex availability.

The rollout also depends on account and workspace settings. Some eligible subscribers may not see the models immediately after the announcement.

OpenAI recommends checking the correct model selector and updating the relevant application if the models are missing.

The availability of GPT-6.2 in Work and Codex should not be interpreted as unlimited usage. Model consumption, credits and speed options can vary by subscription.

GPT-6.2 Introduces Ultrafast Access for More Pro Subscribers

OpenAI is also expanding its Ultrafast processing option with the new release.

Pro 500 subscribers can use Ultrafast with both GPT-6.2 Astra and GPT-6.2 Sol.

For the first time, OpenAI is also making Ultrafast available to Pro 200 subscribers, beginning with GPT-6.2 Sol.

Ultrafast is intended to reduce waiting time for supported model requests, which may be useful for developers and professionals handling time-sensitive assignments.

However, the feature is not included for every subscriber, and its billing rules differ from ordinary model usage.

The expansion shows that OpenAI is placing greater emphasis on performance options for customers who use its models for demanding work.

For many users, the practical benefit will depend on whether faster processing saves enough time to justify any additional usage costs.

How to Use GPT-6.2 Astra and Sol in ChatGPT Work

How to Use GPT-6.2 Astra and Sol in ChatGPT Work

Eligible subscribers can access the models through ChatGPT Work by opening the Work experience and selecting the appropriate model.

The Work model picker allows users to choose GPT-6.2 Astra or GPT-6.2 Sol when those models are available for their account.

Users who prefer the coding environment can select either model through Codex.

In managed Enterprise or Edu workspaces, administrators may need to enable access before the models appear.

OpenAI also recommends updating the ChatGPT desktop application and restarting it if the new models are missing.

For users working through Codex CLI, version 0.166.0 or later is required to access GPT-6.2 Astra and Sol.

If the models remain unavailable after updating, the account may still be waiting for rollout access or may have restrictions associated with its subscription.

Switching back to regular ChatGPT conversations will not make the new models appear in the ordinary Chat model picker.

Developers Can Access GPT-6.2 Through the OpenAI API

The launch also makes both models available to developers through OpenAI’s API.

Applications can request GPT-6.2 Astra using the model identifier gpt-6.2-astra and GPT-6.2 Sol using gpt-6.2-sol.

Both models support text and image inputs and produce text outputs through supported API interfaces.

Developers can use them to build coding assistants, document analysis tools, research applications and other systems that require advanced reasoning or multistep execution.

OpenAI’s model release notes confirm support through the Responses API and Chat Completions API.

API access is billed separately from ordinary ChatGPT subscriptions when developers use their own API credentials.

That means paying for ChatGPT Plus or Pro does not automatically provide unlimited API usage.

Developers should review the relevant pricing and model documentation before moving an existing application to GPT-6.2.

GPT-6.2 Astra and Sol Come With Tighter Cybersecurity Safeguards

OpenAI has also introduced stricter safeguards for some cybersecurity-related requests.

The company says GPT-6.2 Astra has stronger capabilities on cybersecurity evaluations, including tasks involving software vulnerabilities and technical security analysis.

As models become more capable in this area, they may also become more useful for harmful activities such as exploiting systems without authorization.

OpenAI says it has applied tighter restrictions to reduce the risk of misuse.

However, these protections can also affect legitimate security research, including authorized vulnerability testing and defensive analysis.

The company is working on access arrangements for approved security professionals through its Daybreak programs.

These restrictions are relevant to developers who plan to use GPT-6.2 for security testing because some requests may receive different responses from earlier models.

Organizations should evaluate whether the new safeguards affect their existing workflows before switching models.

Why GPT-6.2 Matters for ChatGPT Users

The release highlights a growing difference between using ChatGPT as a conversational assistant and using it as a platform for completing substantial work.

For everyday users, the immediate change may be limited because regular ChatGPT conversations are retaining their existing model choices.

For professionals using Work or Codex, the launch provides access to newer models with improved capabilities and lower costs for certain assignments.

Developers may also benefit from lower flagship API pricing and a cheaper alternative in GPT-6.2 Sol.

However, the value of the upgrade will depend on how the models perform in actual projects.

Users who primarily ask questions, write short messages or seek everyday assistance may not need to switch to Work simply because GPT-6.2 is newer.

Those managing complex documents, software development or long-running assignments may find the additional capabilities more relevant.

The important point is that access to the latest model family now depends on which ChatGPT experience a user chooses.

OpenAI’s Latest Models Are Here, but ChatGPT Access Is Divided

GPT-6.2 Astra and Sol represent another major update to OpenAI’s model lineup, with improved professional capabilities and lower costs for its flagship offering.

The company has made both models available through ChatGPT Work, Codex and the API, giving eligible subscribers and developers new options for coding, research and complex assignments.

At the same time, OpenAI has chosen not to add either model to regular ChatGPT conversations at launch.

That means users who expect every new OpenAI model to appear in the familiar ChatGPT model picker may be disappointed.

The release also reflects a broader shift in OpenAI’s strategy, with different products offering different models and capabilities depending on the type of work being performed.

GPT-6.2 Astra and Sol are available now, but using them requires the right ChatGPT plan and the right product. For regular ChatGPT conversations, the latest model launch does not automatically mean a new model to select.