After Anthropic released Fable 5.1, its popularity declined, while OpenAI’s GPT-6 Astra large model continued to draw attention. This model boasts strong integration capabilities and high token efficiency, effectively reducing usage costs. Iwata Shota, a visiting professor at JAIST in Japan, agrees with Liang Wenfeng, the founder of DeepSeek, believing that focusing on the development of LLMs is the correct path to AGI. Previously, OpenAI adjusted its strategy due to competition, abandoning areas such as video generation and focusing on core coding capabilities, which led to a return to peak performance. Jensen Huang, CEO of NVIDIA, agrees with this and praises its training efficiency, stating that currently only 100,000 graphics cards are needed for training, and it is expected that this number will increase to 400,000 next time.
After Anthropic released Fable 5.1, its popularity declined, while OpenAI’s GPT-6 Astra large model continued to dominate the scene. Iwata Shota, a visiting professor at JAIST in Japan, agrees with Liang Wenfeng, the founder of DeepSeek, believing that focusing on the development of LLMs is the way to achieve AGI. Astra has strong comprehensive capabilities and high token efficiency, reducing usage costs. OpenAI previously adjusted its strategy due to competition, abandoning areas such as video generation and returning to its core coding capabilities, thereby reaching new heights. NVIDIA CEO Jensen Huang agrees and praises its training efficiency, stating that currently only 100,000 graphics cards are used, and this number will increase to 400,000 next time.