← Back to live feed1 story
Power users of the GPT-6 Astra model can now operate at 'low' or 'medium' reasoning efforts to exceed the performance of GPT-5.6 Sol's 'high' setting. This shift in efficiency makes Astra approximately 50% cheaper per task to run for equivalent output, removing the need for the intensive compute levels previously required by the Sol model to achieve similar results.
Recent system updates for ChatGPT account holders reduce the subscription quota drawn by power users by up to 3-4X on the long tail of usage. These optimizations target the usage patterns of high-frequency users to lower subscription overhead without changing the quality of the model's generated responses.
Sign in to suggest edits
Key sources
- SOURCE@thsottiaux“up to 3-4X less usage being drawn from the subscription”x.com
- SOURCE@thsottiaux“GPT-6 Astra on low performs better than GPT-5.6 Sol on high”x.com
- SUPPORT@thsottiaux“about ~50% cheaper per task at roughly equivalent output”x.com
- SUPPORT@reach_vb“Astra provides much higher performance at lower reasoning efforts compared to Sol”x.com
- SUPPORT@npew“Astra just obliterated our hardest benchmark 👀”x.com
- SUPPORT@rhyssullivan“20 minutes later it was done and worked great”x.com
- SUPPORT@scaling01“it logically (no code) solved ALL of them”x.com
- SUPPORT@htihle“It sets a new individual high score on 6 of the 17 tasks”x.com