- Astra improves professional-task benchmarks by 5.2 points while cutting task cost to under one-third of its predecessor
- Tops AA-3D at 1318 Elo, surpassing Fable 5.2's 1271
- Astra API priced at $8 input and $40 output per million tokens, with cached input down 60%
AI CURATED HOT STORY · 11/18/2026
OpenAI Launches GPT-6.2 Astra and Sol with Top 3D Generation
OpenAI has simultaneously released GPT-6.2 Astra and its lightweight counterpart Sol, with Astra achieving comprehensive upgrades in complex agent tasks, scientific research, and 3D generation, and topping the AA-3D benchmark with 1318 Elo, while API prices drop by 20%. Sol's coding ability approaches Astra's at one-fifth the task cost, and both models are now available on ChatGPT Work, Codex, and the API.
According to QuantumBit, Astra improves professional task evaluations by 5.2 percentage points and reduces task cost to less than one-third of the previous generation; its AA-3D score of 1318 Elo surpasses Fable 5.2's 1271; API pricing is $8 input and $40 output per million tokens, with cached input reduced by 60%. Artificial Analysis independently confirms Astra's first-place ranking in interactive 3D experience generation, winning 139 out of 152 human blind tests, indicating that spatial and interactive content generation is becoming a standardized evaluation dimension.
The dual release marks further progress in model tiering and cost optimization: Astra targets high-end professional scenarios, while Sol covers high-frequency tasks like coding at extremely low cost, potentially accelerating AI adoption in development and enterprise workflows. Meanwhile, interactive 3D generation as a new capability dimension could drive reconstruction of content production tools in gaming, film, and the metaverse.
Hourly heat curve
7 fully observed hoursThe line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.
Coverage timeline
Total 2 reports · Latest first- 量子位官网T2·88 pts
- Artificial Analysis (@ArtificialAnlys)T1·58 pts
- GPT-6.2 Astra ranks #1 on the AA-3D interactive 3D experience benchmark
- It won 139 of 152 human head-to-head comparisons, a clear preference margin
- Interactive 3D generation is emerging as a new axis of model evaluation