OpenAI's latest model GPT-6 Astra proves that the company's technology is still on track, which is also good news for Oracle and SoftBank.

Author｜Adam Levin

Editor｜Wang Mei

Last week,OpenAI released its latest, most powerfulAI model, and the company's president Greg Brockman stated:“Welcome to theAGI era.”

AGI stands for “Artificial General Intelligence,” which has long been the goal ofAI research. I used to define this term as “a machine that can perform at least as well as the best humans on any intellectual task.” However, in the current hype surroundingAI, AGI has lost its original meaning.Astra is indeed powerful, but it is still far from being able to replace humans.

I have decided to officially remove the termAGI from my vocabulary, along with its synonym “superintelligence” (superintelligence) — Meta platform loves to use this term.

Aside from Brockman's high-profile statement, his acclaim for this AI laboratory's latest achievements is not without reason.The frontier model at the level of Astra, like Anthropic's Fable, has significant upgrades over existing models. After years of slow, gradual improvements, this generation of models has finally changed the game.Compared to OpenAI and Anthropic's previous top models Sol and Opus, they have achieved a substantial leap, but the costs have also significantly increased. Clients were already deterred by Sol and Opus’s hefty bills, and the new model's fees have doubled.

However, there will always be clients willing to pay. Overall, the agents built on Astra and Fable can collaborate on long-cycle, complex tasks, which is crucial for enterprise scenario implementation. Although the current price may be prohibitively high, it is not hard to imagine that in the future, such AIs will be much cheaper and more widespread.

Just past the long weekend, many tested Astra's capabilities, and many were impressed by its performance. This release is significant for OpenAI, as in the high-value, high-end large model market, OpenAI's market share is being continuously eroded by Anthropic. Although OpenAI led early on, Anthropic's monthly revenue may have already surpassed OpenAI's this summer, and it seems that Anthropic will be the first to go public, with its valuation reportedly reaching $2 trillion.

OpenAI may not go public until next year. However, the launch of Astra is good news, indicating that its technology roadmap is progressing smoothly and should alleviate investors' concerns.Astra's release is also a positive signal for a range of publicly traded companies — these companies, betting on OpenAI becoming a dominant tech platform in the future, have effectively become alternative targets to OpenAI.

SoftBank is OpenAI's largest investor and plans to become an important cloud service provider for OpenAI through its subsidiary SB Energy. SB Energy submitted an IPO prospectus last week, and the funds raised will be used to build a super-large data center for OpenAI.

There is also Oracle, which is already one of OpenAI's cloud service providers. Oracle has $638 billion in backlog orders, of which $300 billion comes from OpenAI contracts, with this order revenue concentrated in the latter half of a five-year cooperation period.

In recent months, as market sentiment towards OpenAI has cooled, both SoftBank and Oracle have had a tough time. Astra may turn the tide. Notably, Brockman stated that Astra was trained on a massive cluster of 100,000 NVIDIA GPUs located in Oracle's data center park in Abilene, Texas. NVIDIA's CEO Jensen Huang mentioned that the next generation of models will require four times as many chips. Those who once questioned why such unimaginable scales of data centers were being built now have their answer.

This Tuesday, SoftBank’s over-the-counter American Depository receipts soared 17.7%, while Oracle's stock price rose 2.4%.

The new generation of models is doing some astounding things. Since Microsoft acquired access to Mythos (a more powerful version than Fable), the number of Windows security fixes released each month during its “Patch Tuesday” has significantly increased. Prior to using Mythos, Microsoft published about 100 fixes monthly; by September, this number reached 974. Oracle patched 1,448 vulnerabilities in its quarterly security updates in July, up from the 481 in April.

Anthropic and OpenAI’s new models help companies like Microsoft discover vulnerabilities in their software at a much faster pace.

These models are also addressing some mathematical problems that have troubled mathematicians for years. Astra seems particularly strong in manipulating software interfaces and automating tasks at high speeds, which has been a challenging area for AI in the past.

The two companies are also feeling the competitive pressure from low-cost models, OpenAI has already lowered product pricing this summer. But Astra and Fable still maintain their leading positions in high-difficulty AI task areas.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。