OpenAI Launches GPT-6.2 Astra and Sol, Cutting Astra API Pricing by 20%
OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, two closed models targeting professional and agentic workloads. GPT-6.2 Astra is the new frontier tier for hard, long-horizon tasks, while GPT-6.2 Sol serves as the efficient everyday tier. For developers, the headline is cost per task: Astra's token prices fall 20%, and OpenAI says both models complete tasks using fewer output tokens. Both are live in the OpenAI API as gpt-6.2-astra and gpt-6.2-sol.
TL;DR
- Size: 1,050,000-token context window, with 922,000 max input and 128,000 max output.
- Runs on: OpenAI's hosted API, ChatGPT Work, and Codex only.
- Performance: Astra scores 62 on the Artificial Analysis Intelligence Index, second behind Claude Fable 5.2 (65).
- Best result: Astra ranks #1 on AA-3D with 1,318 Elo, ahead of Fable 5.2 (1,271).
- Worst result: Astra posts a 59% AA-Omniscience hallucination rate (Sol: 55%), and Sol's Coding Agent Index drops 3 points.
- Bottom line: Frontier-level results at a much lower cost per task, but no coding gains and hallucination remains high.
What Is GPT-6.2 Astra?
Astra is positioned as OpenAI's top-tier reasoning model for complex, multi-step work. Its 1,050,000-token context window and 128,000-token maximum output make it suitable for long documents, extended agent runs, and repository-scale codebases. The 20% reduction in token prices is aimed squarely at teams measuring spend per completed task rather than per token.
Astra is available only through OpenAI's hosted API, ChatGPT Work, and Codex, keeping weights closed and deployment centralized.
What Is GPT-6.2 Sol?
Sol is the efficient everyday tier: a lower-cost default for routine production traffic, lighter agent workflows, and high-volume applications. Like Astra, it benefits from reduced output token usage, though OpenAI reports that Sol's Coding Agent Index slipped 3 points, meaning coding-focused teams should benchmark before switching.
Performance and Benchmarks
On the Artificial Analysis Intelligence Index, Astra reaches 62, placing it second behind Claude Fable 5.2 at 65. Astra leads the AA-3D benchmark with 1,318 Elo versus Fable 5.2's 1,271, suggesting a strong advantage on spatial and 3D reasoning tasks.
The weaker spot is factual reliability. Astra records a 59% AA-Omniscience hallucination rate, with Sol at 55%—figures that remain high for production use and argue for retrieval grounding and verification layers in agentic pipelines.
Key Takeaways for Developers
- Cost efficiency: A 20% price cut plus fewer output tokens lowers cost per task, not just cost per token.
- Choose Astra for depth, Sol for volume: Reserve Astra for long-horizon reasoning and complex agent runs; use Sol for routine traffic.
- Validate coding workloads: With no coding gains reported and Sol's Coding Agent Index down 3 points, run your own evals before migrating.
- Mitigate hallucination: At 59% (Astra) and 55% (Sol) on AA-Omniscience, pair both models with retrieval and output checks in production.
Availability
Both models are accessible today via the OpenAI API as gpt-6.2-astra and gpt-6.2-sol, and through ChatGPT Work and Codex. As closed models, they are not available for self-hosting or local deployment.
via MarkTechPost
