• Deutsch
  • English
  • OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol for different levels of professional work. Astra targets complex projects, while Sol is intended for faster everyday tasks; according to the provider, both are available in ChatGPT Work, Codex, and through the application programming interface (API). For you, the main points are lower usage costs, new specialist tools for sales and law, and substantial remaining limits in reliability and coding performance.

    What are GPT-6.2 Astra and Sol?

    GPT-6.2 Astra and Sol are closed language models, meaning systems that process text and image inputs and return text without making their underlying technology publicly accessible. OpenAI describes Astra as its frontier model for especially difficult, multistep projects. Sol is the more efficient option for everyday work and replaces GPT-6.1 Sol.

    The OpenAI announcement claims better professional output, coding, computer use, and efficiency. According to MarkTechPost’s account, this includes cleaner presentations, spreadsheets, and documents that follow templates and professional conventions more closely. Most of these quality claims come from the provider and have not been independently verified across every use case.

    MarkTechPost reports that both models have a 1.05-million-token context window. A token is a small unit of text used to measure and bill model inputs and outputs. Up to 922,000 tokens can be used for input and 128,000 for output. In principle, this capacity supports very large document collections or long work sessions, but it does not ensure that every detail will be handled reliably.

    The sources do not fully agree about availability. MarkTechPost lists only the API, ChatGPT Work, and Codex, while THE DECODER also names Plus, Pro, Enterprise, and Edu users. Pro-500 subscribers are said to receive Ultrafast mode with both models, while Pro-200 subscribers initially get it with Sol. The supplied sources provide no information about regional restrictions in Switzerland.

    How much do the two models cost?

    The quoted prices cover API use, not a ChatGPT subscription. GPT-6.2 Astra costs $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. Cached inputs are previously processed portions of text that can be billed at a lower rate when reused.

    That makes Astra 20 percent cheaper than GPT-6 Astra, which cost $10 for input and $50 for output per million tokens. Simon Willison points out that the cached-input price falls even more sharply, from $1 to $0.40 per million tokens. For long, recurring work sessions that repeatedly use the same material, this difference may matter more than the general price reduction.

    GPT-6.2 Sol remains at $2 per million input tokens, $0.10 for cached input, and $10 per million output tokens. Sol therefore costs one quarter of Astra’s rate in all three categories and stays at the same level as GPT-6.1 Sol and GPT-6 Sol. The price had already been cut in half from GPT-5.6 Sol, which cost $4 for input and $20 for output.

    For most users, token rates reveal only part of the final cost. Your bill depends on document length, the number of responses, repeated inputs, and the chosen model. OpenAI also says that both models complete tasks with fewer output tokens, but the supplied material does not show whether that claim will reduce the total cost of your particular workflow. Prices for the referenced ChatGPT subscriptions are not included in the reports.

    Which professional applications are being added?

    The clearest expansion involves specialized workflows. With ChatGPT for Sales, OpenAI is introducing a sales plugin and more than 30 new or expanded partner plugins. A plugin connects ChatGPT to a particular function or an external workplace tool.

    According to the provider, sales teams can use the system to research prospects, prepare presentations, and update a customer relationship management (CRM) system, the central platform where a company manages contacts and customer activity. One concrete example is preparing for a sales meeting: ChatGPT can collect information about the target company and turn it into a draft pitch. A second is follow-up work, where meeting details are transferred to the CRM and next steps are prepared.

    Legal work gets a separate product called Astra for Law. OpenAI says the specialized model ranked first in a launch evaluation covering 24 areas of law. Every ChatGPT Enterprise customer can request access. However, the short announcement does not identify the competing models, evaluation method, or specific error rates, so it does not establish general legal reliability.

    The focus on advanced professional work also has a recent precedent. One day before GPT-6.2 launched, THE DECODER reported that the earlier GPT-6 Astra had played a leading role in a possible mathematical proof related to a lattice formulation of the Standard Model of particle physics. The model implemented an idea from physicist Nathaniel Craig mathematically, while other physicists raised critical questions. Considerable work was still required for realistic simulations. The case illustrates specialist potential and the value of human review, but it cannot automatically be treated as evidence about GPT-6.2.

    For Switzerland, the sources do not address support for national languages, local data storage, or specific privacy arrangements. Companies, schools, and public bodies therefore cannot determine from these reports whether sensitive Swiss data is treated differently in the specialist products than in the general services. Availability for Enterprise and Edu accounts is relevant, but it does not replace a review of the applicable contracts and privacy requirements.

    Pros and Cons of GPT-6.2 Astra and Sol

    Pros:

    • Clear division of work – Astra targets difficult, long-running projects, while Sol is the lower-cost choice for frequent everyday tasks.
    • Lower Astra pricing – API input and output rates are 20 percent below the direct predecessor, with a larger reduction for cached input.
    • Large working context – The 1.05-million-token context window can accommodate extensive documents and longer sessions.
    • Professional specialization – Sales features and Astra for Law provide tailored workflows rather than only a general chat window.

    Cons:

    • High error score in one test – MarkTechPost reports 59 percent for Astra and 55 percent for Sol on AA-Omniscience; this describes that benchmark, not the probability that any given answer will be wrong.
    • Little coding progress – Reports describe coding performance as stagnant for both models, and Sol lost three points on the cited Coding Agent Index.
    • Incomplete product details – Subscription prices and Switzerland-specific information about language, availability, and privacy are missing.
    • Limited transparency – Several performance claims come from OpenAI, while specialist evaluations such as Astra for Law are only partly documented.

    The broader comparison results are mixed as well. According to MarkTechPost, Astra scores 62 on the Artificial Analysis Intelligence Index, placing it behind Claude Fable 5.2 at 65. In the spatial AA-3D test, however, Astra leads with an Elo score of 1,318 versus 1,271 for Fable 5.2; Elo is a relative rating system based on direct comparisons.

    A practical experiment by Simon Willison produced similarly mixed results. Astra and Sol were asked in Codex to create a 3D animation in which Yonkers turns into a large machine and moves through New York. In his assessment, neither model handled the central transformation as well as Claude Fable 5.2 or Gemini 4 Carbon. Strong overall scores do not guarantee that an unusual creative assignment will be completed convincingly.

    What does this mean for your work?

    If you are getting started, Sol is the more obvious first test for recurring office work. You could ask it to summarize an existing document or create a presentation draft from material you have already verified. Avoid sensitive data until the privacy conditions for your account and employer are clear, and check names, numbers, and professional claims against the original documents. A large context window does not eliminate fabricated or overlooked details.

    Advanced users may find Astra more useful when a task involves several steps, extensive source material, or strict formatting. For example, you could have it develop a presentation and spreadsheet from the same verified document collection, then inspect whether it followed templates and field conventions. Repeated inputs may benefit from the lower cached-input price, while Sol remains substantially less expensive for preparation and frequent standard tasks.

    Sales teams can begin with a limited workflow: research and draft the pitch first, require human review, and only then update the CRM. In legal work, Astra for Law should be treated as an assistant for reviewing and organizing material, not as a substitute for qualified legal advice. Its reported first-place result across 24 areas of law is a useful signal, but without published test details it is not enough to establish safe performance in a particular case.

    GPT-6.2 Astra and Sol combine lower API costs with a clearer separation between complex professional work and lower-cost everyday use. The new sales and legal tools make their workplace role more concrete, but external testing still points to stagnant coding results and serious problems with incorrect answers. The main open issues are reliability outside selected evaluations and the terms that will apply to Swiss organizations for privacy, language support, and availability.

    Sources

    AI-FunghiAI-Funghi

    Image: Kampus Production via Pexels

    © 2024 - 2026 ai-funghi.com | All Rights Reserved | Impressum | Datenschutz