BlockBeats News, August 18, according to MaxForAI, today, the AI project J-Space Cognition Suite that went viral on the X platform has been questioned by the community, with its promoted DeepSeek V4 test results being unreproducible. The project previously claimed that V4 Flash combined with J-Space could match GLM-5.3, while V4 Pro could outperform Fable 5 on multiple Agent Benchmarks, with a 2.53x speed improvement and a 2.21x Token efficiency boost.

GitHub user GoForceX used 87 questions from Terminal Bench 2.1 for high-concurrency retesting and confirmed the loading of J-Space-related modules. The results showed that after integrating J-Space, the benchmark performance slightly decreased, and both Token usage and costs increased, contrary to the project's claims of performance, speed, and Token efficiency improvements.

Subsequently, the community requested the project to publicly release full evaluation configurations, per-question results, run logs, raw execution times, and Token consumption data. Currently, the project has mainly released summarized results and has not provided complete original experimental records substantial enough to verify the precise data mentioned above. Of note, in response to the skepticism, the project's authors admitted that the related data was indeed "exaggerated," stating that the actual improvements were roughly between 1.6x to 3x. As of now, the authors have not issued a formal response to the community's questions, and some relevant skeptical Issues have been deleted.