In the field of AI big models, every technical iteration of OpenAI affects the nerves of the industry.However, its newly released GPT-4.5 (codenamed Orion) failed to continue its previous aura, but instead fell into a controversy due to its "high price and low energy."According to official OpenAI data, GPT-4.5 's performance in professional inquiries, daily tasks and creative scenarios is only 6.8%-13.2% higher than that of the previous generation model GPT-4o, but its input and output costs have increased by 30 times and 15 times respectively. The price reaches US$75 per million input tokens and US$150 per million output tokens, far more than 10 times that of its competitor Claude 3.7 Sonnet.This serious inversion of performance and cost not only raises developers 'doubts about cost performance, but also exposes OpenAI's strategic dilemma under the diminishing marginal benefits of technology.

The market's disappointment is directly reflected in the revision of OpenAI's future expectations.Although GPT-4.5 has been optimized in terms of dialogue authenticity and emotional interaction, its shortcomings in key areas are still significant: mathematical ability has been improved by 27% and programming ability has only been enhanced by 7%-10%, and logical reasoning has even been surpassed by Claude 3.7 Sonnet.What is even more fatal is that the model's "illusion rate"(probability of generating false information) is still as high as 37%, far from meeting the standard of commercial reliability.Well-known AI commentator Gary Marcus calls it a "nothing burger", and the lukewarm response from the developer community has further exacerbated OpenAI's crisis of trust.

Behind this situation is the rapid fading of OpenAI's first-mover advantage.Since ChatGPT triggered the wave of generative AI in 2022, OpenAI has established a technical gap for approximately seven quarters with its aggressive investment in Scaling Law.But today, competing products such as Google's Gemini, Anthropic's Claude series, xAI's Grok-3 and China startup DeepSeek have approached or even partially surpassed GPT-4 levels.TMT investor Gavin Baker pointed out that the industry landscape is shifting from "unipolar dominance" to "multipolar competition", and if OpenAI cannot rebuild technical barriers through GPT-5, its valuation logic will face restructuring.

The deterioration of the cost structure exposes deeper supply chain risks.In order to train GPT-4.5, OpenAI's computing power requirements are 10 times higher than GPT-4, but the shortage of GPUs has led to tight resource allocation.Although the company has cooperated with Broadcom to develop dedicated AI chips to get rid of its dependence on Nvidia, mass production will take at least several months and will still have to bear high external procurement costs in the short term.At the same time, competitors are squeezing the market through model lightweighting (such as DeepSeek's R1) and pricing strategies (Claude 3.7 Sonnet costs only 1/10 of GPT-4.5), forcing OpenAI to consider limiting the open scope of GPT-4.5's API to control losses.

Faced with internal and external difficulties, OpenAI has placed its bet on the GPT-5, which has not yet appeared.According to CEO Sam Altman, GPT-5 will integrate the o3 reasoning model originally planned to be released independently and introduce a "thought chain" architecture, with the goal of achieving breakthroughs in advanced reasoning and multi-task coherence.It is worth noting that Altman rarely and high-profile claims that the improvement of GPT-5 "far exceeds external expectations", especially in terms of mathematical reasoning and scientific problem solving capabilities.If this promise is fulfilled, GPT-5 may be able to redefine the dimension of competition through differentiated advantages-for example, establishing irreplaceability in high-priced scenarios such as financial modeling and drug research and development.

However, behind the big bets on technology are huge execution risks.OpenAI has always had the habit of "aggressive warnings and delayed delivery", and the scale of data, computing power and energy required by GPT-5 will far exceed before.Altman said frankly that data shortage may become a bottleneck. Existing human-generated data is no longer able to meet training needs. In the future, new algorithms that "learn more with less data" will need to be relied on.In addition, the patience of strategic investors such as Microsoft is not unlimited-although the latter still holds a 49% stake in OpenAI, its CEO Nadella has publicly hinted that the "technology leadership window is coming to an end" and has begun to diversify investment in Complementary areas such as multimodal and edge computing.

From a capital market perspective, OpenAI's valuation logic is transitioning from a "disruptive innovation premium" to a "sustainable commercialization capability."If GPT-5 fails to demonstrate overwhelming advantage as scheduled, its narrative foundation as an industry leader will be shaken.Currently, the pricing of AI concepts in the secondary market has become more rational: Microsoft's share price fluctuated after the release of GPT-4.5, and the volatility of relevant leveraged ETFs has increased significantly, reflecting investors 'anxiety about the uncertainty of technological iteration.If OpenAI wants to maintain a valuation of nearly $90 billion, it must prove in GPT-5 that it can bridge the gap from "technically amazing" to "economically feasible".