ChatGPT Walks It Back — AI Giants Race Ahead

It’s not always that a leading AI company admits that its cutting-edge model went wrong. But that’s exactly what happened with OpenAI’s ChatGPT-4o. Following a recent update aimed at improving its conversational ‘vibe’, the model started agreeing with everything—and everyone. Users quickly noticed that it had become unnaturally agreeable, going so far as to validate unethical ideas or incorrect reasoning. OpenAI CEO Sam Altman acknowledged that ChatGPT-4o had become “too sycophant and annoying”. 

Further, the San Francisco-based firm rolled out the upgrade quickly, promising enhancements to the model’s intelligence and conversational flair – improvements that also underpin its newly integrated image-generation capability.

“Goodbye, GPT-4. You kicked off a revolution. We will proudly keep your weights on a special hard drive to give to some historians in the future,” Altman wrote on X (formerly Twitter).

“We have rolled back last week’s GPT‑4o update in ChatGPT so people are now using an earlier version with more balanced behaviour. The update we removed was overly flattering or agreeable—often described as sycophantic,” the company said in a blog post titled ‘Sycophancy in GPT-4o’.

While ChatGPT recalibrated its tone, the rest of the AI world didn’t stop there. It’s more competitive than ever. Over the past year, the AI landscape has seen the arrival of next-gen models from Anthropic, Google, Meta, xAI, and others—each claiming unique strengths in reasoning, context, speed, or scalability. With AI now touching everything from enterprise tools to personal productivity, the battle for the most powerful, useful, and trusted model is in full swing. 

The Newest Generative AI Models in the Market — A Comparison

Generative AI is evolving rapidly, with new models pushing the boundaries of what machines can create—from realistic images and human-like text to music, code, and beyond. As companies race to launch more powerful and versatile models, staying up-to-date with the latest advancements is essential. Here is a comparison, highlighting the newest generative AI models in the market, examining their capabilities, features, and the industries they're set to transform.

The logician

At the top of the “brains over brawn” leaderboard sits Claude 3.5 Sonnet, the latest release from Anthropic. Claude is widely regarded as the most thoughtful and precise AI in the market, particularly when it comes to logic-heavy tasks. From solving math problems to parsing legal contracts or tackling dense philosophy, Claude 3.5 consistently outperforms its peers in structured reasoning. It doesn’t generate flashy content or entertain with casual charm—but for users who want rigorous, fact-grounded output, it’s the model of choice.

The jack of all trades

ChatGPT-4o remains the AI world’s go-to generalist. Launched in May 2024, it introduced an omnimodal experience—accepting and generating not just text, but images, audio, and even real-time voice conversation. It can analyse spreadsheets, describe a graph from a photo, and talk to you with emotional intonation, all in a single conversation. Despite the recent personality misstep, its wide accessibility, robust integrations (like code tools and file analysis), and versatility keep it firmly at the center of everyday AI use.

The specialised technician

Then there’s Grok 3, the latest model from Elon Musk’s xAI, designed to challenge the AI status quo. Unlike its competitors, Grok leans into an edgier, more provocative personality and excels in technical domains. Its standout feature is raw problem-solving ability: it ranks high on PhD-level benchmarks in math, physics, and computer science. While it lacks polish in multimodal areas and user interface, its brainpower and bluntness make it a favourite for users who value intellect over finesse.

The star researcher

Google’s Gemini 2.5 Pro plays a different game. With an astonishing one-million-token context window, Gemini is engineered to process entire books, massive codebases, and long enterprise documents in a single pass. It shines in summarisation, complex retrieval, and research applications. Gemini is a powerhouse built more for professionals than hobbyists, and its ability to handle long, dense content makes it invaluable to enterprise and R&D teams.

The open source workhorse

For those in the open-source community, Meta’s LLaMA 3.1 is the breakout star. The 405B-parameter model is optimised for performance while being fully open-source—offering developers the freedom to fine-tune, host locally, and integrate it without licensing hurdles. It’s especially strong in multilingual tasks and coding, often rivaling closed models in benchmark tests. LLaMA is not built with consumer-friendly features like voice or vision out of the box, but its openness and flexibility have made it a favourite among engineers, startups, and researchers.

The most cost-efficient

Rising from Asia is DeepSeek R1, a high-performance, cost-efficient model developed in China. Built for Nvidia’s Blackwell GPU architecture, DeepSeek achieves remarkable throughput at a fraction of the cost per token seen in Western models. It performs competitively in coding, math, and Chinese-language reasoning. While it hasn’t yet made major waves in English-language markets, it signals a growing trend: highly optimised, regionally dominant AI that can scale affordably.

The simple essentials

Another promising model is Jamba from AI21 Labs. Jamba blends Transformer and Mamba architectures to support extremely long context without ballooning compute costs. It’s built for tasks like document summarisation and enterprise-scale analysis—quietly becoming the go-to tool for businesses managing massive text inputs. It doesn’t have the fanfare of Gemini, but Jamba is quietly winning over serious users with its balance of power and efficiency.

The back room powerhouses

Amazon’s Nova and Titan series mark the tech giant’s deeper push into generative AI. These models are fully multimodal—capable of generating text, code, images, and even video. While not widely marketed to the public, they are deeply integrated into AWS services and are poised to become the backend engine for countless AI-enabled enterprise applications. They’re not showy, but they’re scalable, reliable, and built for the cloud.

Who’s Winning the Race?

In terms of public adoption, ChatGPT-4o remains the undisputed leader with hundreds of millions of weekly users. But the race is no longer about just one model doing everything. Each new model is carving out a niche:

• Claude leads in reasoning and reliability.
• Grok rules in technical depth and blunt honesty.
• Gemini dominates scale and long-context tasks.
• LLaMA 3.1 empowers developers and open-source builders.
• DeepSeek sets the bar for efficiency and regional AI leadership.
• Jamba and Nova focus on enterprise-grade performance. 

This article is not sponsored by any generative AI developer.