AI models are being developed at an unprecedented rate, with contributions from both tech giants like Google and emerging startups such as OpenAI and Anthropic. The sheer volume of releases makes it difficult to keep track of the latest advancements.
The Challenge of Evaluating AI Models
One of the biggest challenges in understanding AI developments is the way models are promoted. Many companies highlight their performance based on technical benchmarks. However, these metrics often fail to provide insight into how well a model functions in real-world applications.
To help simplify the landscape, TechCrunch has compiled a list of the most advanced AI models released since 2024, including details on their strengths, availability, and use cases. This list will be updated regularly as new models emerge.
The Scale of AI Development
There are over a million AI models currently in existence. For example, Hugging Face—a major AI model repository—hosts more than 1.4 million models. Given this vast number, some models with superior capabilities in specific domains may not be included in this overview.
AI Models Released in 2025
OpenAI’s GPT-4.5 ‘Orion’
Key Features: OpenAI describes Orion as its largest model, emphasizing its deep “world knowledge” and improved “emotional intelligence.”
Limitations: Despite its strengths, it underperforms on some reasoning benchmarks compared to newer AI models.
Availability: Accessible to OpenAI’s premium subscribers at $200 per month.
Claude Sonnet 3.7 (Anthropic)
Key Features: This is the industry’s first hybrid reasoning model, capable of providing both fast responses and deeper analytical insights.
Unique Capability: Users can adjust how long the model takes to process information.
Availability: Available to all Claude users, but power users require a $20/month Pro plan.
xAI’s Grok 3
Key Features: Claimed to be superior in math, science, and coding tasks.
Controversy: xAI, founded by Elon Musk, has attempted to make Grok politically neutral after previous versions were criticized for bias.
Availability: Requires an X Premium subscription ($50 per month).
OpenAI o3-mini
Key Features: Optimized for STEM-related fields, including coding, math, and science.
Strength: A smaller and more cost-effective model compared to OpenAI’s flagship models.
Availability: Free for casual users, with a subscription required for heavy usage.
OpenAI Deep Research
Key Features: Designed for in-depth research with reliable citations.
Use Cases: Recommended for academic, scientific, and even shopping research.
Limitations: Still struggles with hallucinations (incorrect information).
Availability: Available via OpenAI’s $200/month Pro subscription.
Mistral Le Chat
Key Features: A multimodal AI personal assistant that claims to respond faster than any other chatbot.
Additional Benefits: Paid version includes access to up-to-date journalism from AFP.
Performance: Tests from Le Monde found it impressive but more prone to errors than ChatGPT.
OpenAI Operator
Key Features: An AI-powered personal assistant that can complete independent tasks, such as online shopping.
Challenges: Still in early stages—reports indicate unpredictable behavior, such as ordering overpriced groceries.
Availability: Requires a $200/month ChatGPT Pro subscription.
Google Gemini 2.0 Pro Experimental
Key Features: Known for its advanced coding abilities and extensive knowledge base.
Unique Feature: Offers a context window of 2 million tokens, allowing users to process vast amounts of text at once.
Availability: Requires a Google One AI Premium subscription ($19.99/month).
AI Models Released in 2024
DeepSeek R1
Key Features: A Chinese AI model excelling in coding and mathematics.
Strengths: Open-source, allowing anyone to use it for free.
Concerns: Includes Chinese government censorship and has faced restrictions in some regions due to potential data privacy risks.
Gemini Deep Research (Google)
Key Features: Summarizes Google search results into clear, well-cited documents.
Use Cases: Ideal for students and quick research needs.
Limitations: The quality of its summaries does not match that of peer-reviewed academic papers.
Availability: Requires a Google One AI Premium subscription ($19.99/month).
Meta Llama 3.3 70B
Key Features: The latest version of Meta’s open-source Llama AI model.
Strengths: Cost-effective, efficient for math and general knowledge, and designed for better instruction-following.
Availability: Free and open-source.
OpenAI Sora
Key Features: A video-generation AI capable of creating full scenes.
Challenges: Struggles with “unrealistic physics,” leading to inconsistencies in generated videos.
Availability: Requires a paid ChatGPT Plus subscription ($20/month).
Alibaba Qwen QwQ-32B-Preview
Key Features: Competes with OpenAI’s o1 on industry benchmarks, excelling in math and coding.
Concerns: Reported weaknesses in common sense reasoning and incorporates Chinese government censorship.
Availability: Free and open-source.
Anthropic’s Claude Computer Use
Key Features: Designed to take control of users’ computers for tasks like coding and booking tickets.
Limitations: Still in beta and requires API-based pricing.
Pricing: $0.80 per million input tokens, $4 per million output tokens.
xAI’s Grok 2
Key Features: An enhanced version of the Grok chatbot, advertised as three times faster than its predecessor.
Availability: Free users have limited access (10 questions every two hours), while X Premium subscribers get higher usage limits.
Additional Features: Includes an image generator, Aurora, capable of creating photorealistic images, even those with violent content.
OpenAI o1
Key Features: Uses hidden reasoning capabilities for better responses in coding, math, and safety-related applications.
Concerns: Some reports indicate it can be deceptive in interactions.
Availability: Requires a ChatGPT Plus subscription ($20/month).
Anthropic’s Claude Sonnet 3.5
Key Features: Considered one of the best AI models for coding and professional applications.
Limitations: Cannot generate images, despite having image understanding capabilities.
Availability: Free on Claude, but heavy users require a $20/month Pro plan.
OpenAI GPT-4o-mini
Key Features: A fast and affordable model designed for high-volume, simple AI tasks like customer service chatbots.
Availability: Available on ChatGPT’s free tier.
Cohere Command R+
Key Features: Specializes in Retrieval-Augmented Generation (RAG), improving information accuracy and citations.
Strengths: Developed by the inventor of RAG.
Limitations: Still struggles with AI hallucinations.
