OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gains
First reported by The Decoder ·
Developers can now access more powerful AI models at a lower cost, potentially reducing the expense of building AI-powered applications.
OpenAI has launched two new large language models, GPT-6.2 Astra and GPT-6.2 Sol, offering significant performance improvements at reduced API prices. Astra is tailored for specialized professional tasks in fields like science, law, and finance, while Sol is designed for quick, everyday applications. Both models are now available to various subscription tiers and via the OpenAI API. GPT-6.2 Astra is priced 20% lower than its predecessor, with specific rates for input, cached input, and output tokens. OpenAI highlights Astra's strong performance on benchmarks like Agents' Last Exam and Harvey LAB-AA, noting its superior cost-performance ratio compared to competitors such as Anthropic's Claude Fable 5.2 and Google's Gemini 4 Carbon. GPT-6.2 Sol maintains its predecessor's pricing and offers competitive performance for general tasks. Both models show advancements in areas like coding, cybersecurity, and 3D design, though some benchmarks indicate areas for further improvement, particularly in coding and hallucination rates.
The release of GPT-6.2 Astra and Sol signals a continued push by major AI providers to balance performance gains with economic accessibility. OpenAI's aggressive pricing strategy, particularly the 20% reduction for Astra's API, directly challenges competitors and may force others to re-evaluate their cost structures. This move could democratize access to advanced AI capabilities, enabling smaller businesses and independent developers to integrate sophisticated models into their products without prohibitive costs.
The specific targeting of Astra for professional work, evidenced by its performance on specialized benchmarks, suggests a market segmentation strategy where highly capable, domain-specific models will coexist with general-purpose ones. Developers should monitor how these specialized models integrate into existing workflows and what new applications they enable. The reported improvements in areas like cybersecurity also indicate a growing emphasis on safety and robustness, though OpenAI's introduction of stricter safeguards may present new challenges for legitimate security research.
AI-written summary. May contain errors.