The key facts
- OpenAI released GPT-6.2 Astra (for complex tasks) and Sol (a general-purpose model)
- Astra scored 77.6–78.3% on Agents' Last Exam versus 73.1% for GPT-6 Astra at roughly half the cost
- The models are available to Plus, Pro, Enterprise, and Edu subscribers; API pricing is $8 and $2 per 1 million input tokens respectively
Two new flagship models
OpenAI has introduced the artificial intelligence models GPT-6.2 Astra and Sol, as announced on the company's website. Astra is positioned as an advanced model for especially complex tasks, while Sol is a universal option for everyday work. According to the developers, both models handle complex tasks better while consuming fewer resources.
Benchmark results and quality of work
OpenAI cites the Agents' Last Exam benchmark for multi-step tasks as the main measurable confirmation of the improvements: GPT-6.2 Astra scored 77.6% at a cost of $4.47 at the 'maximum' effort level and 78.3% at $2.89 at 'very high'. For comparison, GPT-6 Astra, released in September, scored 73.1% at a cost of $8.99 — meaning the new model simultaneously improved the result and reduced the cost of running the test roughly two- to threefold.
What else the developers highlight
The company highlights Astra's improved handling of content: documents, spreadsheets, and slides, in OpenAI's wording, 'look less like AI-generated material', and the program code has become 'cleaner'. The model's strong results in science benchmarks, modeling, and cybersecurity are noted separately. Sol, according to its creators, allows more tasks to be completed within the plan's usage limit compared to its predecessors and features a 'clear and natural' communication style.
Availability and pricing
Both models are already available to Plus, Pro, Enterprise, and Edu subscribers. On the Pro 500 plan costing $500 per month, the 'ultrafast' Ultrafast mode works with any model; on Pro 200 at $200, it works only with Sol. API pricing for GPT-6.2 Astra: $8 per 1 million input tokens, $0.4 per 1 million cached input tokens, and $40 per 1 million output tokens; Sol's rates are five times lower across the board — $2, $0.1, and $10 respectively.

