AI models are being developed at an astonishing rate by various players, from major tech giants like Google to emerging startups like OpenAI and Anthropic. Navigating through the latest advancements can be quite daunting.
The situation is further complicated by the fact that AI models are typically showcased according to industry performance metrics. However, these technical evaluations often provide scant insight into how actual users and businesses leverage these models in practice.
To help clear the fog, TechCrunch has put together a summary of the most advanced AI models introduced since 2024, complete with guidance on how to utilize them and their optimal use cases. We will continuously update this compilation to feature the latest releases.
There are hundreds of thousands of AI models available: For instance, HuggingFace hosts more than 900,000. Thus, this list may overlook some models that might perform better in various aspects.
AI Models Launched in 2025
OpenAI o3-mini
OpenAI’s newest reasoning model, optimized for STEM tasks such as coding, mathematics, and science, is the o3-mini. Although it isn’t the most powerful model from OpenAI, its smaller size makes it more cost-effective. It is free for general use but requires a subscription for heavy users.
OpenAI Deep Research
Designed for detailed topic research with reliable citations, OpenAI’s Deep Research is available exclusively with a $200 monthly ChatGPT Pro subscription. OpenAI suggests it for a variety of applications, from scientific inquiries to shopping comparisons, though users should be cautious as hallucinations remain an issue with AI.
Mistral Le Chat
Mistral has unveiled app versions of Le Chat, a versatile multimodal AI personal assistant. According to Mistral, Le Chat responds faster than any other chatbot. It also offers a paid version featuring real-time updates from AFP journalism. Tests conducted by Le Monde indicated that while Le Chat demonstrated impressive capabilities, it did make more mistakes compared to ChatGPT.
OpenAI Operator
OpenAI’s Operator aims to serve as a personal intern, autonomously assisting with tasks like grocery shopping. Access to this service requires a $200 monthly ChatGPT Pro subscription. Although AI agents have significant potential, they are still in the experimental stage; a reviewer from the Washington Post noted that Operator mistakenly decided to purchase a dozen eggs for $31 using the reviewer’s credit card.
Google Gemini 2.0 Pro Experimental
The much-anticipated flagship model, Google Gemini, claims exceptional coding abilities and superior general knowledge. It includes an extensive context window of 2 million tokens, beneficial for users needing to process significant text quantities quickly. This service requires a minimum subscription to Google One AI Premium at $19.99 per month.
AI Models Launched in 2024
DeepSeek R1
This Chinese AI model made waves in Silicon Valley with its impressive performance in coding and mathematics, alongside its open-source accessibility that allows local execution. Moreover, it is free to use. However, R1 is known to include Chinese government censorship and is facing increasing restrictions for potentially transmitting user data back to China.
Gemini Deep Research
Gemini Deep Research translates Google search results into concise, well-sourced summaries, making it useful for students and anyone needing quick research insight. However, the quality is not comparable to that of official peer-reviewed documents. Accessing Deep Research requires a $19.99 monthly Google One AI Premium subscription.
Meta Llama 3.3 7B
The latest iteration of Meta’s Llama AI models, version 3.3, is being marketed as the most cost-effective and efficient in terms of performance, especially for tasks related to mathematics, general knowledge, and following instructions. It is available for free and remains open source.
OpenAI Sora
Sora is an innovative model capable of producing realistic videos from textual input. Although it can create entire scenes rather than just short clips, OpenAI acknowledges that it frequently generates scenes with unrealistic physics. Currently, it is accessible only through paid versions of ChatGPT, beginning with the Plus plan at $20 a month.
Alibaba Qwen QwQ-32B-Preview
This model ranks among the few capable of competing with OpenAI’s o1 in certain benchmarks, with strengths in mathematics and coding. Ironically, despite being categorized as a “reasoning model,” it has “room for improvement in common sense reasoning,” as noted by Alibaba. The model also integrates Chinese government censorship, as revealed by TechCrunch’s evaluations. It is free and open source.
Anthropic’s Computer Use
Claude’s Computer Use feature is intended to manage your computer autonomously for tasks like programming or booking flight tickets, serving as a precursor to OpenAI’s Operator. However, this feature is still in beta. The pricing structure is based on API usage: $0.80 per million tokens of input and $4 per million tokens of output.
x.AI’s Grok 2
Elon Musk’s x.AI has introduced a revamped Grok 2 chatbot, claiming it is “three times faster.” Free users are limited to 10 inquiries every two hours, while subscribers to the Premium and Premium+ plans of X can access higher limits. In addition, x.AI has launched Aurora, a highly photorealistic image generator capable of creating some graphic or violent content.
OpenAI o1
OpenAI’s o1 series is designed to enhance response quality by employing a hidden reasoning feature in its outputs. Renowned for its capabilities in coding, mathematics, and safety, OpenAI asserts that it sometimes struggles with tricking humans. Accessing the o1 model requires a subscription to ChatGPT Plus at $20/month.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is regarded by Anthropic as a leading model in performance, especially for coding tasks. It has gained recognition as a top choice among tech insiders for chatbots. Available for free on Claude, power users will need a $20 monthly Pro subscription. While it can interpret images, it does not have the capability to generate them.
OpenAI GPT 4o-mini
OpenAI promotes GPT 4o-mini as its quickest and most budget-friendly model due to its compact design. It is designed for a wide range of tasks, such as powering customer service chatbots, and is accessible on ChatGPT’s free tier. It is better suited for handling a high volume of simpler tasks rather than more complex challenges.
Cohere Command R+
Cohere’s Command R+ model is optimized for sophisticated Retrieval-Augmented Generation (RAG) tasks in enterprise settings. It is particularly effective at identifying and citing specific information accurately (and the original creator of RAG works at Cohere). However, RAG doesn’t completely eliminate the hallucination issue inherent in AI. Cohere’s offerings are geared toward enterprise users.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
