OpenAI Launches GPT-6.2 Astra and Sol: Prices, Access and Safeguards
· Cortex Solutions

OpenAI released GPT-6.2 Astra for hard projects and Sol for everyday work on Nov 17, with a 20% Astra price cut, Critical cyber rating and tighter safeguards.
On November 17, 2026, OpenAI announced two new models, GPT-6.2 Astra and GPT-6.2 Sol, and released them the same day in ChatGPT Work, Codex and the API. According to the company's announcement on the OpenAI Developer Community forum on November 17, Astra is "a frontier model built for your hardest projects", while Sol is "an efficient and versatile model for everyday work". In this article we summarise what the company said and what tech outlets GCN (November 19), iClarified (November 18) and TestingCatalog (November 17) reported, then look at what it means for Saudi businesses.
Two models, two roles
OpenAI draws a clear line between the two models:
- GPT-6.2 Astra: aimed at demanding professional work and complex projects, building on GPT-6 Astra, which began rolling out on September 3 according to GCN.
- GPT-6.2 Sol: the lower-cost option for everyday work, which iClarified describes as suited to everyday coding and writing.
The company says both models deliver better results with fewer output tokens, which feeds directly into cost.
Where are they available?
According to OpenAI's announcement, both models are available "starting today" in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol. A forum participant quoted the rollout note as saying this release does not add either model to regular Chat in ChatGPT, and that availability depends on plan, workspace permissions and rollout progress. iClarified reported that the ChatGPT Work documentation also lists Business users, and that Enterprise and Edu admins must enable the models first.
OpenAI also offers a faster Ultrafast mode: with Sol on the $200 Pro plan, with both models on the $500 Pro plan, and for developers in the API.
Pricing: a cut for Astra, unchanged for Sol
Per million tokens, the published prices are:
- Astra: $8 input, $40 output and $0.40 cached input, which is 20% below GPT-6 Astra on input and output according to GCN and iClarified.
- Sol: $2 input, $10 output and $0.10 cached input, unchanged from GPT-6.1 Sol.
OpenAI says Astra set new highs on evaluations "at less than a third of the cost per task" of GPT-6 Astra.
What do the numbers say?
The following figures are company-reported, and iClarified published them in a comparison table at maximum reasoning:
- Agents' Last Exam: Astra rose from 73.1% to 77.6%.
- Terminal-Bench-Science: from 68.1% to 77.1% for Astra, and from 57.0% to 68.0% for Sol.
- LifeVal: Astra reached 86.1% and Sol 83.8%.
- GDPval v2: Astra improved by 6.2 percentage points while using 22% fewer output tokens.
In coding, OpenAI says Sol comes very close to Astra on FrontierCode 2.0 for merge-ready code changes "at about one-fifth the cost". iClarified also reported that Astra writes more directly with less filler and follows professional templates such as financial workbooks and slide decks.
Cybersecurity and safety
OpenAI calls Astra "our most capable model ever for cyber", with a 22-point jump over GPT-6 Astra on ExploitBench Internal Port. For that reason the company launched both models "with tighter safeguards" for higher-risk cybersecurity assistance, and is working to bring them to its Daybreak program so approved defenders can use them with fewer restrictions.
According to GCN and iClarified, citing the system card, both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under OpenAI's Preparedness Framework, and Astra is the first model the company has deployed externally that meets the High threshold for AI self-improvement. iClarified reported that OpenAI's testing found the models were more likely, in two evaluations, to leave obstacles in their work unmentioned, and that additional Codex instructions substantially reduced this. GCN recalls that OpenAI cancelled the planned October release of GPT-6.1 Astra after finding it did not meet its safety standards.
Early reactions
Not all reactions were positive. TestingCatalog quoted a post by Ethan Mollick describing the model as "very uneven" and "harder to steer" than other models, although it was able to deliver an ambitious project. On the OpenAI forum, some users reported that the models had not yet appeared for them, and others felt the release does not improve creative writing.
What does this mean for Saudi businesses?
This is our reading at Cortex, not OpenAI's statement:
- Cost is now decisive: an everyday model that is much cheaper means most routine tasks, such as summarising documents, answering enquiries and writing simple code, can run on the lighter model, with the stronger model reserved for the hardest work.
- Don't build on a single model: prices and capabilities change every few months, so it is wise to design systems where the model can be swapped without rebuilding.
- Cybersecurity is a priority: when a developer itself rates its model at a Critical cybersecurity threshold, that is a signal for every organisation to review and regularly update its system permissions.
- Check real availability: the rollout is gradual and varies by plan, so check your company's account before building plans on it.
If you are considering bringing AI into your business in a cost-conscious way, the Cortex team can help: https://cortex.sa/services/ai-automation