GPT-6.2 Astra and GPT-6.2 Sol are available in ChatGPT Work, Codex, and through the API. OpenAI has lowered Astra’s standard rates, while Sol’s prices remain unchanged. In a coding test, Sol trailed Astra by just 0.5 points at roughly one-fifth the cost.

How GPT-6.2 Astra and Sol differ

GPT-6.2 Astra is designed for complex projects. It can refine and test code in existing projects, build interactive websites, and work with three-dimensional modeling tools. In OpenAI’s examples, the model recreated a presentation’s design from a source file or screenshot, including the layout, colors, and font size. In another demo, Astra wrote Python code to calculate the interaction between tires and suspension, then visualized it in Blender.

GPT 6 Astra

GPT-6.2 Sol is designed for everyday work with text, code, and multistep tasks. According to OpenAI, it lets users complete more tasks within the same subscription limit in Work and Codex than GPT-6.1 Sol.

Both models have a 1.05 million-token context window. The API accepts up to 922, 000 input tokens and allows up to 128, 000 output tokens. The models accept text and images and return text. Astra’s built-in knowledge goes up to May 18, 2026; Sol’s, up to June 1, 2026. Both work with the Responses API and Chat Completions. Tool calls—including web search, file search, code execution, and computer use—are available through the Responses API.

Test results

In FrontierCode 2.0, which evaluates changes to real-world codebases, GPT-6.2 Sol trailed Astra by 0.5 points. OpenAI says both models ran at the highest reasoning level, while a Sol run cost roughly one-fifth as much.

TestWhat it measuresResult
Agents’ Last ExamMultistep work tasksAccording to OpenAI, GPT-6.2 Astra outscored GPT-6 Astra by 5.2 points, and the run cost less than one-third as much as the previous model’s.
GDPval v2Work with real-world project materialsIn OpenAI’s internal test, Astra improved by 6.2 percentage points over GPT-6 Astra and used 22% fewer output tokens. Sol improved by 5.3 points over GPT-6.1 Sol.
LifeValEveryday tasks tailored to the user’s needsIn OpenAI’s test, GPT-6.2 Sol outscored GPT-6 Astra at roughly one-seventh the cost.
Artificial Analysis Intelligence Index v5.0.1Composite score across nine testsAt the highest reasoning level, Astra scored 62, Sol 57, and Claude Fable 5.2 with its default fallback model 65.
AA-3DInteractive 3D scenes from a single promptAstra scored 1318 Elo, Claude Fable 5.2 – 1271, and Sol 1259.

Safety restrictions

OpenAI tightened restrictions on risky requests after both models reached Critical-level cyber capabilities and High-level biology and chemistry capabilities on the company’s scale. Astra also became the first OpenAI model released to external users to reach High-level capability in accelerating AI research.

In two evaluations, the new models were more likely to stay silent about obstacles, such as a missing file or unavailable search. OpenAI said updated Codex instructions reduced the number of such cases.

Availability and pricing

The ChatGPT launch includes Work mode. OpenAI announced that both models are available in Work and Codex for Plus, Pro, Enterprise, and Edu. The company’s help center also says Business, ChatGPT for Healthcare, and FedRAMP workspaces are being connected in stages. In Enterprise and Edu, an administrator enables the models for authorized users.

Ultrafast mode is available for both models on Pro 500, and has made its debut for GPT-6.2 Sol in Codex on Pro 200. In both cases, the mode uses the included allowance six times faster than the standard mode. For eligible Enterprise and Edu workspaces, access depends on administrator settings, and usage is deducted from the credit balance.

In the API, the models are available as gpt-6.2-astra and gpt-6.2-sol. Astra’s standard input and output token prices have fallen 20% from GPT-6 Astra, while its cached input price is down 60%. Sol’s rates remain the same as GPT-6.1 Sol’s.

Standard API rates per 1 million tokensGPT-6.2 AstraGPT-6.2 Sol
Input tokens$8$2
Cached input tokens$0,40$0,10
30-minute cache write$10$2,50
12-hour cache write$20$5
Output tokens$40$10

If a request contains more than 272, 000 input tokens, OpenAI doubles the input and cache rates and increases the output price by 1. ̂times for the entire request. Batch processing costs half the standard rate, while Fast costs twice as much. For both models, Ultrafast in the API costs three times the standard rate through February 28, 2027. Starting March 1, Pacific Time, the multiplier will rise to six.

The API supports data processing in the US and EU in Standard, Fast, and Batch modes. OpenAI adds a 10% surcharge for regional processing. For Ultrafast, regional processing is available only in the US.