GPT-4.5 "both poor and expensive," if GPT-5 is not released soon, OpenAI will face difficulties

Leading the AI wave with its GPT series models, OpenAI's recently launched GPT-4.5 has encountered a setback.

This new model, which was highly anticipated, not only underperformed expectations but also came with an astonishingly high price. The industry generally believes that OpenAI is facing unprecedented pressure, and if it cannot quickly launch a groundbreaking GPT-5, its leading position may be at risk.

High Price, Low Performance: OpenAI Lowers Expectations

The competition for AI large models is becoming increasingly intense, and last week OpenAI launched the GPT-4.5 (Orion) large model.

However, unlike before, OpenAI has deliberately lowered market expectations for this new model. OpenAI CEO Sam Altman stated on social media platform X that GPT-4.5 is a "huge, expensive model" and "will not achieve overwhelming advantages in benchmark tests."

OpenAI's own assessment also shows that GPT-4.5 lags behind Anthropic's latest release, Claude 3.7 Sonnet, in multiple metrics, and even falls short of OpenAI's own reasoning model launched months ago.

Even more astonishing is that GPT-4.5 is priced at $75 per million input tokens and $150 per million output tokens, which is 30 times that of GPT-4o and over 10 times that of Claude 3.7 Sonnet.

Although GPT-4.5 has improved in terms of conversational authenticity and humor, making it potentially more suitable for voice interactions, its overall performance clearly did not meet OpenAI's initial expectations. The Information pointed out that this may be evidence of diminishing returns in pre-training.

Developers' reactions to GPT-4.5 have generally been lukewarm. Notable AI commentator Gary Marcus bluntly called GPT-4.5 a "nothingburger." Ars Technica quoted an anonymous expert saying that GPT-4.5 is "flashy but impractical," and its performance does not justify such a high price.

Even in the area of "emotional intelligence," which OpenAI prides itself on, GPT-4.5 has failed to gain widespread recognition. Many users have joked on social media that while GPT-4.5 has high emotional intelligence, it is not very good at work; what they need is a smart model, not one that can understand their feelings.

Even more concerning is that GPT-4.5 still exhibits hallucinations and errors. According to MIT Technology Review, OpenAI itself has admitted that GPT-4.5 has a 37% chance of "fabricating facts" in a key test.

OpenAI's First-Mover Advantage Fades

If OpenAI cannot quickly launch a significantly improved model, doubts about its weakening leading position will grow louder.

Previously, renowned TMT investor Gavin Baker stated in a tweet that the transformation of the AI industry landscape is accelerating, and OpenAI's first-mover advantage is fading, with Microsoft also choosing to step back.

Gavin pointed out in the tweet:

When ChatGPT burst onto the scene in November 2022, OpenAI established a dominant position in the generative AI field for seven consecutive quarters through aggressive bets on Scaling Law. However, this advantage window is closing: Google's Gemini, xAI's Grok-3, and Deepseek's latest models have all reached a technical level comparable to GPT-4.

Even OpenAI founder Altman has pointed out that OpenAI's future leading advantage will be narrower; Microsoft CEO Nadella essentially stated that their unique period of leading model capabilities is coming to an end.

Faced with the dilemma of GPT-4.5, the industry is turning its attention to OpenAI's next-generation model GPT-5 and its o3 reasoning model. However, OpenAI has a habit of "previewing" new technologies, and the actual release time is often delayed.

It has been over two months since OpenAI announced the performance evaluation of o3, but Sam Altman recently stated that o3 will not be released as an independent model like o1.

Although Gavin remains optimistic about OpenAI, he predicts that in the future, data will become the core of competition, and cutting-edge models that cannot obtain unique and valuable data will be the fastest depreciating assets in history.

In the increasingly fierce competition of AI large models, relying solely on pre-training has become insufficient to establish a long-term advantage. Whether OpenAI can return to the peak with GPT-5 will depend on its breakthroughs in data acquisition, reasoning capabilities, and technological innovation