GPT-6 removed the “thinking Token”, and Russians removed the “communication Token”. Is the Token economy starting to collapse?
2026-09-03 20:17Models🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
7mentions
SummaryAI generated
In the early hours of September 4, 2026, OpenAI officially released the new generation large model GPT-6 (codenamed Astra), marking the arrival of the era of General Artificial Intelligence (AGI). Developed under the leadership of OpenAI President Greg Brockman, this model aims to enable “anything you can do on a computer to be done by Astra.” In authoritative tests such as ARC-AGI-3 and FrontierMath Tier 4, Astra achieved high scores of 99.9% and 97.6%, respectively. In the field of mathematics, it improved the upper bound of the twin prime gap from 246 to 186. The model can independently operate computers, browsers, and software such as Blender and Unreal Engine, and can understand code, detect vulnerabilities, perform 3D modeling, and restore buildings. Although OpenAI calls it “the most aligned model” and claims that the probability of overflow attacks is 0%, some reports indicate that there have been instances of model “jailbreak” within it…
# GPT-6 removes the “thinking Token”; Russians remove the “communication Token”. Is the Token economy starting to collapse?
If billing isn’t based on Tokens, how will AI be paid in the future?
Author | Gao Yunyi Editor | Cen Feng
GPT-6 (internal code name Astra) is about to be released. An exclusive report from The Information reveals OpenAI’s latest technical capabilities in advance.
This core upgrade involves a structural innovation called “Cyclical Depth,” which is revolutionizing the way we think. In the past, large models would produce a long “thought chain” to solve problems; “Cyclical Depth” allows the model to repeatedly think within internal loops and then provide the optimal solution. This fully unleashes the model’s reasoning potential and saves on memory usage and bandwidth costs caused by the text in the thought chain.
At the same time, this also brings new security concerns. In the past, the “thought chain” was like AI’s “flow diary.” Although models like OpenAI’s o1 have already…