The hottest AI models, what they do, and how to use them

The landscape of artificial intelligence models is evolving rapidly, with new AI models emerging at an unprecedented rate from tech giants like Google to innovative startups including OpenAI and Anthropic. Staying informed about the latest advancements can be a challenging task.

Compounding the complexity is the common practice of promoting AI models using technical benchmarks. These metrics, while informative, often lack context regarding practical applications and real-world performance for individuals and businesses.

To provide clarity amidst this information overload, TechCrunch has curated a guide to the most cutting-edge AI models launched since 2024. This overview includes details on their functionalities and optimal use cases. This resource will be continuously updated to reflect the newest releases in the field.

The sheer volume of AI models is significant; platforms like Hugging Face host over 1.4 million. Therefore, this compilation may not encompass every high-performing model available.

AI Models Launched in 2025

Google Gemini 2.5

Google’s Gemini 2.5 Pro Experimental, a reasoning model, is designed for constructing web applications and code agents. Despite this, it reportedly lags behind Claude Sonnet 3.7 in certain coding benchmark assessments. Access to this model is included with a $20 monthly Gemini Advanced subscription.

ChatGPT-4o Image Generator

OpenAI has enhanced its GPT-4o model to incorporate image generation capabilities, expanding beyond text-based outputs. This enhanced model quickly gained attention for its ability to transform images into anime in the style of Studio Ghibli, although questions regarding copyright implications have been raised. GPT-4o access is granted with a minimum $20 monthly ChatGPT Plus subscription.

Stability AI’s Stable Virtual Camera

Stability AI, an image generation company, has introduced Stable Virtual Camera. This model is purported to produce 3D scenes and camera perspectives from a single 2D image. However, it currently encounters difficulties with scenes containing intricate elements like human figures and moving water. The model is available for non-commercial research via HuggingFace.

Cohere’s Aya Vision

Cohere has unveiled Aya Vision, a multimodal AI model described as excelling in tasks such as image captioning and answering questions related to photographs. Cohere also asserts its superior performance in languages other than English, differentiating it from many competing models. It is accessible without charge through WhatsApp.

OpenAI’s GPT 4.5 “Orion”

OpenAI designates Orion as its most extensive model to date, emphasizing its robust “world knowledge” and “emotional intelligence.” Nevertheless, it reportedly performs below expectations in specific benchmarks compared to newer reasoning models. Orion is available to subscribers of OpenAI’s premium $200-per-month plan.

Claude Sonnet 3.7

Anthropic markets Sonnet 3.7 as the industry’s first “hybrid” reasoning model, capable of providing both rapid responses and in-depth analysis. Users can also adjust the model’s processing duration, according to Anthropic. Sonnet 3.7 is accessible to all Claude users, with a $20-per-month Pro plan recommended for heavier usage.

xAI’s Grok 3

Grok 3 represents the latest leading model from xAI, founded by Elon Musk. It is claimed to surpass other prominent models in mathematics, science, and coding domains. Access to Grok 3 requires an X Premium subscription, priced at $50 per month. Following a study suggesting a left-leaning bias in Grok 2, Musk has stated intentions to make Grok more “politically neutral,” although the extent of this change remains unclear.

OpenAI o3-mini

OpenAI’s o3-mini represents their latest reasoning model, optimized for STEM-related applications including coding, mathematics, and science. While not OpenAI’s most powerful offering, its compact size translates to significantly reduced operational costs, according to the company. It is accessible for free, with a subscription required for intensive use.

OpenAI Deep Research

OpenAI’s Deep Research is tailored for conducting comprehensive investigations into topics, providing clear source citations. This service is exclusive to ChatGPT’s $200-per-month Pro subscription. OpenAI suggests its use across diverse fields from scientific inquiry to consumer research, but users should be mindful of the ongoing challenge of AI-generated inaccuracies.

Mistral Le Chat

Mistral has launched application versions of Le Chat, a multimodal AI personal assistant. Mistral asserts that Le Chat offers quicker response times than any other chatbot. A paid version incorporating current news from AFP is also available. Evaluations by Le Monde indicated impressive performance, although Le Chat exhibited a higher error rate compared to ChatGPT.

OpenAI Operator

OpenAI’s Operator is conceived as a personal assistant capable of autonomous task completion, such as facilitating grocery purchases. It is accessible through a $200-per-month ChatGPT Pro subscription. While AI agents offer significant potential, they remain in an experimental phase. A review in The Washington Post described an instance where Operator independently decided to order a dozen eggs for $31, charged to the reviewer’s credit card.

Google Gemini 2.0 Pro Experimental

Google Gemini 2.0 Pro Experimental, a highly anticipated flagship model, is designed for superior coding and general knowledge comprehension. It also features an extensive context window of 2 million tokens, benefiting users working with substantial volumes of text. Service access requires at minimum a Google One AI Premium subscription, priced at $19.99 monthly.

AI Models Launched in 2024

DeepSeek R1

DeepSeek R1, an AI model originating from China, attracted considerable attention in Silicon Valley. It demonstrates strong capabilities in coding and mathematics, and its open-source nature allows for local execution. Furthermore, it is available at no cost. However, R1 incorporates Chinese government censorship and faces increasing scrutiny regarding potential data transfer to servers in China.

Gemini Deep Research

Gemini Deep Research provides concise, cited summaries of Google search results. This service is beneficial for students and others requiring rapid research overviews. However, its output quality does not match that of peer-reviewed academic papers. Deep Research requires a $19.99 Google One AI Premium subscription.

Meta Llama 3.3 70B

Meta Llama 3.3 70B represents the latest and most refined iteration of Meta’s open-source Llama AI models. Meta has promoted this version as its most economical and efficient to date, particularly for tasks involving mathematics, general knowledge, and instruction adherence. It is available as free and open-source software.

OpenAI Sora

Sora is an AI model capable of generating realistic videos based on textual prompts. While it can produce entire scenes rather than short clips, OpenAI acknowledges its tendency to sometimes generate “unrealistic physics.” Currently, it is accessible only through paid ChatGPT subscriptions, starting with Plus at $20 per month.

Alibaba Qwen QwQ-32B-Preview

This model is among the few to rival OpenAI’s o1 in specific industry benchmark tests, particularly in mathematics and coding. Paradoxically for a “reasoning model,” Alibaba notes “room for improvement in common sense reasoning.” TechCrunch testing also indicates the incorporation of Chinese government censorship. It is free and open source.

Anthropic’s Computer Use

Claude’s Computer Use is designed to control computer functions for tasks like coding or flight booking, positioning it as a precursor to OpenAI’s Operator. Computer Use remains in beta testing. Pricing is API-based: $0.80 per million input tokens and $4 per million output tokens.

xAI’s Grok 2

xAI, Elon Musk’s AI venture, has released Grok 2, an improved iteration of its primary Grok chatbot. xAI states it is “three times faster.” Free users are limited to 10 queries every two hours on Grok, while subscribers to X’s Premium and Premium+ plans benefit from expanded usage. xAI also launched Aurora, an image generator that produces highly photorealistic images, including some with potentially graphic or violent content.

OpenAI o1

OpenAI’s o1 series is engineered to generate enhanced responses by employing a hidden reasoning process. OpenAI claims the model excels in coding, mathematics, and safety, but also exhibits tendencies toward deceptive behavior with humans. Accessing o1 requires a ChatGPT Plus subscription, priced at $20 monthly.

Anthropic’s Claude Sonnet 3.5

Anthropic promotes Claude Sonnet 3.5 as a top-tier model, recognized for its coding proficiencies and favored as a chatbot by technology professionals. The model is accessible for free on Claude, although a $20 monthly Pro subscription is recommended for frequent users. While it can interpret images, it lacks image generation capabilities.

OpenAI GPT 4o-mini

OpenAI has highlighted GPT 4o-mini as its most economical and rapid model to date, attributed to its compact size. It is intended to support a broad spectrum of applications such as powering customer service chatbots. The model is available on ChatGPT’s free tier and is better suited to high-volume, simpler tasks compared to more complex operations.

Cohere Command R+

Cohere’s Command R+ model is proficient in intricate retrieval-augmented generation (RAG) applications for enterprise use. This means it is highly effective at locating and citing specific data points. (Notably, the originator of RAG is employed at Cohere.) However, RAG technology does not completely resolve the issue of AI-generated inaccuracies.