AiStaffo

OpenAI Releases GPT-6.2 Astra and Sol, Cuts Astra API Price 20%

Stock photo: laptop desk documents

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026. Astra costs $8 per million input tokens and $40 per million output tokens in the API, which is 20% below GPT-6 Astra. Sol keeps its earlier rate of $2 input and $10 output per million tokens. Both models are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. OpenAI's own safety documentation rates both models at its Critical cybersecurity threshold. For businesses that automate work through the API, the lower Astra rate changes the cost of each request, but the total bill still depends on how many tokens each job uses.

In short

  • OpenAI released GPT-6.2 Astra and Sol on 17 November 2026 in ChatGPT Work, Codex and the API.
  • Astra's API price is $8 input and $40 output per million tokens, 20% below GPT-6 Astra.
  • Sol keeps $2 input and $10 output per million tokens.
  • OpenAI's system card rates both models at the Critical cybersecurity threshold, so access controls matter.
  • A lower token price lowers job cost only if the job does not need more tokens.

What happened

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026. According to OpenAI's developer community announcement, both models are rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and developers can reach them through the OpenAI API as gpt-6.2-astra and gpt-6.2-sol. Crypto Briefing describes Astra as aimed at complex professional, scientific and coding work, and Sol as the more affordable model for everyday tasks.

The price change is the detail most relevant to businesses that run automated work through the API. VentureBeat reported that Astra now costs $8 per million uncached input tokens and $40 per million output tokens, while Sol keeps the $2 input and $10 output rates of its predecessor, GPT-6.1 Sol. Both models have a context window of 1.05 million tokens and support up to 128,000 output tokens, according to the same report.

Model (API name)Input per million tokensOutput per million tokens
GPT-6.2 Astra (gpt-6.2-astra)$8$40
GPT-6.2 Sol (gpt-6.2-sol)$2$10

All figures are in US dollars, as published by OpenAI and reported by VentureBeat on 17 November 2026. Businesses paying in other currencies will see the charge converted by their card provider or billing platform.

Why it matters for businesses that automate work

Most automated office work, such as reading invoices, drafting follow-ups or checking reconciliations, runs on the API rather than in a chat window. A lower price per token therefore reaches the cost of those jobs directly. The 20% cut applies only to Astra. Sol, which is the lower-cost model, did not change.

The safety classification also matters to anyone who gives a model access to company systems. OpenAI's GPT-6.2 system card states that both Astra and Sol meet the Critical threshold for cybersecurity capabilities under the company's Preparedness Framework, and that both meet the High threshold for biological and chemical capabilities. According to OpenAI's launch post, the models ship with tighter safeguards for higher-risk cybersecurity assistance. iClarified reports that Astra is the first externally released OpenAI model rated High for AI self-improvement, a category that measures the ability to speed up AI research.

What changes in practice

For a business that automates work, the practical effect depends on its size and setup. A five-person firm that uses ChatGPT Work through a Plus or Business plan does not pay per token and will see no direct price change. A firm with a development team, or a group that calls the API from its own software, will see the lower Astra rate on every Astra request.

Consider two illustrative cases. A bookkeeping team in Mumbai that sends bank statements through an API script to extract transactions would pay the Astra rate on the input tokens of each statement and the output tokens of each extracted list. A distribution company in Chicago that drafts supplier reminders with a model would see a smaller saving, because short reminders use few output tokens. In both cases the owner should compare the cost of a finished job, not the price of one token.

DataCamp, in its launch review published in 2026, reported a test in which Astra cost $1.79 against $2.93 for GPT-6 Astra on one dashboard task. That is one reviewer's test on one task, and it does not predict the cost of a different workload.

Three limits apply. First, a lower token price does not reduce the number of tokens a task needs, so a poorly written prompt can still cost more. Second, model answers still need a person to check any figure that goes into a payment, a tax return or a customer document. Third, VentureBeat reported that several AI power users criticised the models after testing them before launch, and that OpenAI's Daybreak programme still directs approved security researchers to GPT-6 Astra or GPT-6 Cyber for authorised security work. Businesses should test the new models on their own files before moving a workflow over.

What to watch next

Three things are worth watching over the coming weeks. The first is whether independent benchmark groups confirm OpenAI's claims on professional tasks. The second is whether the Critical cybersecurity rating leads to changes in who can use Astra for security-related work, which the launch materials already hint at through the Daybreak arrangement. The third is whether the lower Astra price holds in later price lists. Teams that pin a model version in their software should check the model name before each update.

How AiStaffo would automate this

For a business that runs document-heavy routine work, AiStaffo would first map which steps call a model through the API and which run inside a chat window, since only API calls carry a per-token charge. The AI worker would be connected to the document inbox, the accounting or ERP system, and the approval queue, so that each request is logged with the model name and token count. Routine extraction and matching would run automatically, while a person still checks any figure that feeds a payment, a tax filing or a customer record. The owner keeps the decision on which model each workflow uses and reviews the monthly token log. To see which of your routine jobs would benefit, book a free automation audit.

Questions people ask

How much does GPT-6.2 Astra cost in the API?
According to VentureBeat, Astra costs $8 per million uncached input tokens and $40 per million output tokens. OpenAI's developer announcement and Crypto Briefing both describe the same rates in US dollars. These are list prices, and the bill for any job depends on how many tokens it uses.
Did GPT-6.2 Sol get a price cut?
No. VentureBeat reported that Sol keeps the $2 input and $10 output rates of GPT-6.1 Sol. Only Astra's API price fell by 20%.
Which plans can use GPT-6.2 Astra and Sol in ChatGPT?
According to OpenAI's developer community announcement, both models are rolling out in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Access through the API is separate and is billed per token.
What does the Critical cybersecurity rating mean for a business?
OpenAI's GPT-6.2 system card says both models meet the Critical threshold for cybersecurity capabilities under its Preparedness Framework. For a business, this means access to security-related use may be restricted, and the company applies extra monitoring to external use. Businesses should limit which staff and systems can call these models.
Is a cheaper model always cheaper for my workflow?
Not always. A lower price per token helps only if the job does not need more output tokens to finish. One reviewer's test on DataCamp showed a lower cost on one task, but that does not predict other workloads. Measure the cost of a completed job before changing a workflow.
Should I switch my automations to GPT-6.2 Astra now?
Test first on a sample of your own files and compare accuracy and cost with your current model. Keep a person checking any output that feeds payments, tax or customer records. A free automation audit can show which of your routine jobs are worth moving.

Book a free automation audit

Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

We use these details only to reply to this request and prepare your audit, and keep them for up to two years. We send no marketing unless you tick the box above. Privacy policy

openaiai modelsai safetybusiness automation