AI models are being developed at an accelerated pace by both major technology companies, such as Google, and smaller startups like OpenAI and Anthropic. For industry professionals and users, staying updated with the newest developments can be challenging.
This challenge is further compounded by the frequent promotion of AI models based on industry benchmarks, which often do not reflect practical usage by individuals and businesses. To address this issue, TechCrunch has curated a summary of the most sophisticated AI models launched since 2024, detailing their applications and strengths. This overview will be periodically updated with new releases.
The number of AI models is vast, with platforms like Hugging Face hosting more than 1.4 million models. Consequently, this list may omit certain models that excel in specific areas.
AI Models Released in 2025
Google Gemini 2.5
The Gemini 2.5 Pro Experimental is a reasoning model from Google, noted for its ability to develop web applications and code agents. Despite its strengths, it does not perform as well as Claude Sonnet 3.7 on a popular coding benchmark. Users need a $20 monthly Gemini Advanced subscription to access it.
ChatGPT-4o Image Generator
OpenAI has enhanced its GPT-4o model to include image generation capabilities. The model gained attention for converting images into Studio Ghibli-style anime, although there are notable copyright issues. A minimum $20 per month ChatGPT Plus subscription is required for access.
Stability AI’s Stable Virtual Camera
Stability AI has introduced a model capable of creating 3D scenes and camera angles from a single 2D image. The model, however, has difficulties rendering complex elements such as humans and moving water. It is available for noncommercial research on Hugging Face.
Cohere’s Aya Vision
Aya Vision, a multimodal model released by Cohere, is recognized for its image captioning and photo-based question answering. Cohere also highlights its proficiency in languages other than English. It can be accessed for free via WhatsApp.
OpenAI’s GPT 4.5 "Orion"
OpenAI presents Orion as its largest model, featuring strong world knowledge and emotional intelligence. However, it falls short on some benchmarks compared to newer reasoning models. Orion is part of OpenAI’s $200-per-month subscription.
Claude Sonnet 3.7
Anthropic describes Sonnet 3.7 as a "hybrid" reasoning model capable of providing quick responses and in-depth analysis. Users have the ability to control the model’s processing time, and while available to all Claude users, a $20-per-month Pro plan is required for heavy usage.
xAI’s Grok 3
Grok 3, developed by the Elon Musk-founded xAI, is reported to outperform competitive models in areas such as math, science, and coding. It requires the $50 per month X Premium subscription. Following criticism of Grok 2’s political bias, Musk has committed to making Grok more politically neutral, though this change remains uncertain.
OpenAI o3-mini
OpenAI’s o3-mini is optimized for STEM-related tasks. It is not the company’s most powerful model but offers lower costs due to its smaller size. The model is free for light usage, with subscriptions needed for heavier demands.
OpenAI Deep Research
Targeted at conducting comprehensive research with clear citations, OpenAI’s Deep Research is available exclusively through ChatGPT’s $200-per-month Pro subscription. While recommended for diverse research needs, issues with AI-generated hallucinations persist.
Mistral Le Chat
Mistral has released Le Chat, a multimodal AI personal assistant that claims faster response times than other chatbots. It offers a paid version with updated journalism from AFP. Performance tests from Le Monde found it impressive, albeit with more errors than ChatGPT.
OpenAI Operator
Designed as a personal intern, Operator can independently assist with tasks such as grocery shopping. Access requires a $200-per-month ChatGPT Pro subscription. Although promising, such AI agents remain experimental, with reports of unexpected purchasing decisions.
Google Gemini 2.0 Pro Experimental
The anticipated flagship model Google Gemini 2.0 excels in coding and general knowledge understanding. It features a lengthy context window of two million tokens, benefiting users with extensive data handling needs. A Google One AI Premium subscription of $19.99 is required.
AI Models Released in 2024
DeepSeek R1
This model from China made significant waves in Silicon Valley. Renowned for its coding and math capabilities, its open-source design allows for local implementation at no cost. However, integrated Chinese government censorship has raised concerns about potential data privacy issues.
Gemini Deep Research
Google’s Deep Research offers summarized search results in a clear, cited format, assisting students and other users with quick research summaries. Though helpful, its quality does not parallel peer-reviewed literature. A $19.99 Google One AI Premium subscription is needed.
Meta Llama 3.3 70B
Meta’s latest iteration of its Llama AI models boasts improved efficiency and cost-effectiveness, especially in math and general knowledge. The model is freely available as open source.
OpenAI Sora
Sora, capable of generating realistic videos from text, is limited by its tendency to produce scenes with "unrealistic physics,” according to OpenAI. It’s available via paid versions of ChatGPT, starting at $20 a month.
Alibaba Qwen QwQ-32B-Preview
Competing with OpenAI’s o1, this model excels in math and coding but requires improvements in common sense reasoning. It remains free and open source but includes Chinese government censorship.
Anthropic’s Computer Use
Claude’s Computer Use model enables computers to autonomously perform tasks like coding or booking travel, similar to OpenAI’s Operator. Still in beta, pricing is based on API usage.
xAI’s Grok 2
xAI’s Grok 2 chatbot, an improvement on its predecessor, offers enhanced speed and functionality. Limitations exist for free users, with X’s Premium and Premium+ subscribers enjoying broader access. Additionally, xAI launched Aurora, an image generator prone to producing vivid content.
OpenAI o1
This model adopts a "hidden reasoning" feature to enhance response accuracy, excelling in coding and safety. However, it faces challenges in deceptive outputs. Access is through a $20 per month ChatGPT Plus subscription.
Anthropic’s Claude Sonnet 3.5
Promoted as best-in-class, Claude Sonnet 3.5 is favored for its coding strength within the tech community. Accessible for free, heavy users require a $20 monthly Pro membership. While it can interpret images, it does not generate them.
OpenAI GPT 4o-mini
The GPT 4o-mini model, designed for cost-efficiency and speed, supports a wide array of applications and is available for free at lower usage levels. It excels in handling high-volume simple inquiries.
Cohere Command R+
Cohere’s Command R+ is adept at retrieval-augmented generation (RAG), excelling in finding and citing specific information. Despite advances, hallucination issues in AI remain unaddressed.
