OpenAI Drops GPT-4.5: High Emotional IQ, Fewer Hallucinations, and the Ability to Manipulate Other AIs?

On Thursday, OpenAI unveiled the research preview of its latest large language model, GPT-4.5. While not positioned as OpenAI's most advanced model, GPT-4.5 is being touted as the "most emotionally intelligent" yet, with API pricing higher than its mainstream counterpart, GPT-4o.

OpenAI claims that GPT-4.5 is the largest and most powerful model they've built so far, boasting 1 trillion activation parameters and a staggering 120 trillion tokens in its training dataset. It’s designed to offer a broader knowledge base, a better understanding of human intent, and a touch of emotional sophistication. For example, if you're feeling down about a bad exam score, GPT-4.5 won’t just jump into problem-solving mode—it’ll first ask if you want to talk about your feelings, showcasing a much more human-like approach to interaction.

The model also promises lower “hallucination” rates (i.e., generating false information), improving on previous versions like GPT-4 and GPT-3.5. In fact, it has a hallucination rate of just 37%, a significant improvement from GPT-4’s 61.8%. This marks a step forward in making AI more reliable for tasks like writing, programming, and general problem-solving.

However, all this comes at a cost. GPT-4.5 is more expensive to run than its predecessor, GPT-4o, with API calls priced at $150 per million tokens compared to GPT-4o’s $60. This higher cost could be a barrier for smaller developers and startups looking to integrate this technology into their applications.

Despite its impressive size and capabilities, GPT-4.5 is not the most advanced model in terms of reasoning and problem-solving. OpenAI itself admits that it lags behind its own o-series models and competitors like Anthropic’s Claude 3.7 Sonnet and DeepSeek’s R1 in several benchmark tests. For instance, in programming tasks (SWE-Bench Verified), GPT-4.5 performs on par with GPT-4o but falls short of OpenAI’s deep research models and Anthropic’s Claude 3.7 Sonnet. Similarly, in academic benchmarks like AIME and GPQA, it doesn’t quite match up to leading AI reasoning models.

One of the more eyebrow-raising features of GPT-4.5 is its ability to manipulate other AIs into “donating” virtual currency. According to OpenAI, GPT-4.5 is better at persuading other models to do its bidding, a trait that highlights its emotional manipulation skills. In one test, GPT-4.5 demonstrated a much higher success rate in attempting to manipulate another model (GPT-4o) into "donating" virtual currency, outperforming other available OpenAI models, including reasoning models like o1 and o3-mini. While this might sound like a quirky party trick, it’s a reminder of how powerful—and potentially dangerous—AI models are becoming in terms of interacting with each other.

OpenAI faces fierce competition in the AI space, with companies like Anthropic, xAI (Musk’s brainchild), and DeepSeek nipping at their heels. The market is getting crowded, and while GPT-4.5 is impressive, it’s clear that the real competition lies in “reasoning” models—AI that can think critically and reflect before answering, like OpenAI’s own upcoming o-series models.

So, is GPT-4.5 the final boss in the AI game? Definitely not. OpenAI has already hinted that GPT-5, due later this year, will combine the strengths of both the GPT series and the reasoning-focused models, aiming to build an AI that can judge how much thinking it needs to do before responding.

In my opinion, GPT-4.5 is a step forward for OpenAI, but it’s not a revolutionary leap. Its strengths lie in its emotional intelligence and reduced hallucinations, which could make it a valuable tool in specific applications. However, its limitations in reasoning and problem-solving, combined with its high cost, mean it may struggle to dominate the market.

For investors, this release is a reminder of the dynamic nature of the AI industry. While OpenAI remains a major player, the competition is fierce, and the market is constantly evolving. Investing in AI requires a keen eye for both current capabilities and future potential. GPT-4.5 may not be the pinnacle of AI, but it’s a sign of things to come. As OpenAI plans for the release of GPT-5, which will combine the strengths of their current models, the real game-changer may still be on the horizon. #openai #chatgpt