GPT-6.2 Astra topped the AA-3D benchmark, which evaluates how well models can create interactive web applications in 3D from text prompts. Human evaluators preferred its results in 139 of 152 comparisons, with a score of 1,318 Elo, 47 points ahead of Claude Fable 5.2 and 59 points ahead of GPT-6.2 Sol. Astra also scored 79 on the AA-Robotics v0.1 test, compared with 78 for Sol and 75 for Claude Fable 5.2. Its API price is set at $8 per million tokens, while Sol reportedly costs about 15% of Astra’s price per task. These results put OpenAI in the lead on 3D experience generation, although they are based on human preferences rather than an absolute measure of performance.
Source: Read the original article


