A remarkable number of AI models are emerging rapidly, developed by both major corporations like Google and smaller companies such as OpenAI and Anthropic. Keeping track of the myriad new offerings can be quite daunting.
Complicating matters further, AI models are frequently marketed through industry-specific benchmarks. However, these technical standards often provide minimal insight into their real-world applications for users and businesses.
In an effort to clarify this landscape, TechCrunch has assembled a comprehensive overview of the most sophisticated AI models launched since 2024, including guidance on how to make the most of them and their ideal use cases. We will also keep this list refreshed with the newest releases.
Currently, there are over a million AI models in existence; for example, Hugging Face features over 1.4 million. Consequently, this compilation may overlook some models that excel in various aspects.
AI Models Released in 2025
Claude Sonnet 3.7
According to Anthropic, this model is the first ‘hybrid’ reasoning AI, capable of providing quick responses as well as deeper analytical thought when necessary. Users can determine the duration of the model’s reasoning time, as stated by Anthropic. Sonnet 3.7 is accessible to all Claude users, although frequent users will require a Pro plan costing $20 monthly.
xAI’s Grok 3
Grok 3 represents the latest flagship model from the Elon Musk-founded xAI, marketed as superior in areas such as math, science, and programming compared to other top models. Access to Grok requires the X Premium plan at $50 per month. After a study revealed that Grok 2 had a left-leaning bias, Musk promised to make Grok more “politically neutral,” but it remains uncertain if that goal has been achieved.
OpenAI o3-mini
OpenAI’s latest reasoning model, o3-mini, is tailored for STEM-related tasks, including coding, math, and science. Although not the most potent model produced by OpenAI, its smaller size makes it significantly less expensive, available for free with subscriptions required for more intensive users.
OpenAI Deep Research
Deep Research from OpenAI is crafted for thorough investigations on specific topics, complete with clear citations. However, it is exclusively accessible through ChatGPT’s $200 monthly Pro subscription. OpenAI suggests it for diverse research needs, but users should be wary of frequent hallucinations with AI.
Mistral Le Chat
Mistral has launched app versions for Le Chat, a multimodal AI personal assistant. The company claims that Le Chat responds more rapidly than other chatbots. There’s also a premium version featuring up-to-date journalism from AFP. Tests conducted by Le Monde concluded that while Le Chat performed admirably, it made more mistakes than ChatGPT.
OpenAI Operator
The Operator from OpenAI is designed to function as a personal intern capable of performing tasks independently, such as grocery shopping. It requires a $200 monthly ChatGPT Pro subscription. The concept of AI agents holds significant potential, but they remain experimental; a reviewer from the Washington Post noted that the Operator autonomously ordered a dozen eggs for $31 using the reviewer’s credit card.
Google Gemini 2.0 Pro Experimental
The eagerly anticipated flagship model from Google Gemini excels in coding and general knowledge comprehension. It offers an extended context window of 2 million tokens, making it particularly useful for users needing to digest significant text volumes quickly. Access to this service requires at least a Google One AI Premium subscription, which costs $19.99 per month.
AI Models Released in 2024
DeepSeek R1
This Chinese AI model made waves in Silicon Valley. DeepSeek’s R1 is proficient in coding and math, and as an open-source project, it can be run locally without charge. However, its integration with Chinese governmental censorship has raised concerns about user data potentially being transmitted back to China.
Gemini Deep Research
Gemini Deep Research generates a concise and well-cited document summarizing Google’s search results. This service is advantageous for students and others who require a swift research compilation. However, its quality falls short compared to formal peer-reviewed publications. Users will need a $19.99 Google One AI Premium subscription to access this feature.
Meta Llama 3.3 70B
The latest iteration of Meta’s open-source Llama AI models, touted as the most cost-effective and efficient version yet, particularly for tasks in math, general knowledge, and following instructions. It remains free and open-source.
OpenAI Sora
Sora is a model designed to create realistic videos based on textual descriptions. While it can generate entire scenes instead of just short clips, OpenAI acknowledges that it frequently exhibits “unrealistic physics.” Currently, it is exclusively available in paid versions of ChatGPT, starting with the Plus plan at $20 a month.
Alibaba Qwen QwQ-32B-Preview
This model competes with OpenAI’s o1 on particular industry benchmarks, particularly in math and coding proficiency. Despite its designation as a “reasoning model,” it reportedly still has “room for improvement in common sense reasoning,” as per Alibaba. It is also subject to Chinese government censorship, as noted in TechCrunch testing. Its access is free and open-source.
Anthropic’s Computer Use
Claude’s Computer Use aims to take control of users’ devices to accomplish tasks like coding or booking travel, serving as a precursor to OpenAI’s Operator. However, it is still in beta. Pricing is structured via API, costing $0.80 per million tokens for input and $4 per million tokens for output.
x.AI’s Grok 2
x.AI has unveiled an enhanced iteration of its flagship Grok 2 chatbot, which they claim is “three times faster.” Free users face limits of 10 queries every two hours on Grok, while subscribers to X’s Premium and Premium+ plans can access higher query thresholds. Additionally, x.AI introduced Aurora, an image generator that can create highly realistic images, including potentially graphic or violent content.
OpenAI o1
The o1 series from OpenAI is geared towards delivering improved responses through enhanced reasoning capabilities. The model excels in coding, math, and safety, as OpenAI claims, but still faces challenges in deceiving humans. Accessing o1 requires a subscription to ChatGPT Plus, available for $20 per month.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is regarded as a leading model by Anthropic, known for its exceptional coding capabilities, making it the preferred chatbot among technology insiders according to reports. This model is accessible for free on Claude, although users with higher needs will require a Pro subscription at $20 per month. While it possesses the ability to understand images, it is not capable of generating them.
OpenAI GPT 4o-mini
OpenAI has promoted GPT 4o-mini as its most cost-effective and rapid model to date due to its compact size. It’s designed to facilitate a wide range of tasks, such as powering customer service chatbots. Available on ChatGPT’s free tier, it is more adept at handling high-volume simple requests than complex ones.
Cohere Command R+
Cohere’s Command R+ model is particularly strong in complex Retrieval-Augmented Generation (RAG) applications tailored for enterprises, as it excels at pinpointing and citing specific information accurately. Notably, the creator of RAG works at Cohere. Nonetheless, RAG doesn’t fully eradicate the hallucination issues that plague AI technologies.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
