By OpenAI has officially launched GPT-4.5, an advanced AI model that the company claims is more emotionally intelligent and less prone to hallucination.
The model, which is now available to ChatGPT Pro users at $200 per month, marks the last iteration of OpenAI’s pretraining approach.
The release has sparked debates on whether the AI leader is losing ground as competitors like Anthropic and DeepSeek introduce models with stronger reasoning capabilities.
GPT-4.5 is significantly more expensive to operate than GPT-4o, with costs ranging from 15 to 30 times higher.
Despite its improvements, OpenAI has remained cautious in its claims, with CEO Sam Altman downplaying expectations by stating that GPT-4.5 is not a “frontier model.”
The company also made a last-minute revision to its technical paper, removing the assertion that GPT-4.5 was not an advanced AI system—an action that raised further speculation.
Meanwhile, Anthropic’s Claude 3.7 Sonnet and China’s DeepSeek R1 are gaining traction, offering more sophisticated reasoning capabilities.
As the AI race intensifies, OpenAI’s dominance appears less certain, especially with GPT-5’s launch on the horizon.
Benchmark data raises doubts
Fortune reports that while OpenAI’s latest model outperforms GPT-4o on several key metrics, it has failed to make a breakthrough in logical reasoning, coding, and multilingual problem-solving.
Early benchmark comparisons indicate that GPT-4.5 struggles against Anthropic’s latest Claude model, which uses a more advanced approach that blends intuitive responses with deep reasoning.
Unlike GPT-4.5, Claude 3.7 Sonnet decides in real time whether to generate instant responses or engage in a complex “chain of thought” process to refine its answers.
This ability makes Claude more adaptable across different types of queries. OpenAI’s latest release lacks this flexibility, raising concerns that its AI models are becoming increasingly outdated in an evolving market.
Social media reactions have been lukewarm, with some AI researchers posting underwhelming benchmark results.
AI expert Gary Marcus described GPT-4.5 as “a nothing burger,” reflecting growing doubts about OpenAI’s ability to maintain its technological edge.
OpenAI shifts AI strategy
GPT-4.5 model, internally referred to as “Orion,” is the last to be built using OpenAI’s long-standing pretraining strategy, which relied on expanding model size and data input.
Going forward, OpenAI is shifting towards reasoning models, which use reinforcement learning to enhance logical processing at test time.
Companies like Anthropic and Google are investing heavily in models that can dynamically adjust their computational resources based on task complexity.
DeepSeek, a rising AI firm in China, has also introduced reasoning-driven models that challenge OpenAI’s current technology.
As competition intensifies, OpenAI is under pressure to deliver a true next-generation model.
CEO Sam Altman has confirmed that GPT-5 will be announced in the coming months, promising a hybrid approach that combines GPT-style fluency with the step-by-step logic of reasoning models.
Whether this shift will restore OpenAI’s lead remains to be seen.
Anthropic, DeepSeek challenge OpenAI
The AI landscape is rapidly evolving, with multiple challengers emerging to disrupt OpenAI’s dominance.
Anthropic has positioned itself as a leader in reasoning AI, while DeepSeek’s R1 model has demonstrated strong results in coding and mathematical reasoning.
Meta and Google, meanwhile, continue to refine their AI offerings, leveraging vast computational resources to push the boundaries of generative AI.
With OpenAI’s technological supremacy now being questioned, the AI industry is entering a new phase where no single company holds a definitive advantage.
As GPT-5’s launch nears, OpenAI faces the challenge of proving that it can keep up with an industry shifting towards reasoning-driven models.
The era of simply scaling up AI is ending, and the companies that adapt to this new reality will define the future of artificial intelligence.
The post OpenAI’s GPT-4.5 launch signals shifting AI race as Anthropic, DeepSeek gain ground appeared first on Invezz
