The AI model is cranked at a pace of dizziness, from everyone in large companies, such as Google, to new companies such as Openai and Anthropic. Tracking the latest products can be overwhelming.
What is added to the confusion is that the AI model is often promoted based on industry benchmarks. However, these technical metrics often say little about how real people and the company actually use.
To reduce noise, TechCrunch compiled the overview of the most advanced AI models launched since 2024 and provided detailed information about how to use and most suitable. We will update this list with the latest launch.
There is literally a million AI model. For example, more than 1.4 million hugs. Therefore, this list can miss some models that are better performed in any way.
The AI model was released in 2025
Cohere’s AYA Vision
Cohere claims that it is best suited to do the same thing, such as the AYA VISION multi -modal model to capture images and answer questions about photography. Also, unlike other models, it is excellent in language other than English. It is provided free of charge in WhatsApp.
Openai’s GPT 4.5 ‘Orion’
Openai called Orion the largest model to date and promoted strong “world knowledge” and “emotional intelligence”. However, compared to the new reasoning model, the performance is low in certain benchmarks. ORENAI offers $ 200 plans to subscribers a month.
Claude Sonnet 3.7
Anthropic says this is the industry’s first ‘hybrid’ reasoning model. Because it can be thought of when you need it when you need it. You can also control the time that users can think of. Sonnet 3.7 is available to all Claude users, but more users require $ 20 a month.
XIA’S GOOK 3
Grok 3 is the latest flagship model of the ELON Musk-Reled Startup XAI. It claims to surpass other major models for mathematics, science and coding. This model requires $ 50 a month. After discovering that Grok 2 was tilted to the left in one study, MUSK promised to change Grok more “politically neutrally,” but it is not yet clear whether it was achieved.
Openai O3-Mini
This is the latest reasoning model of Openai and is optimized for STEM -related work such as coding, mathematics and science. It’s not the most powerful model of Openai, but it’s smaller, so it’s very cheap. It can be used for free, but you need to subscribe for heavy users.
Openai Deep Research
The deep study of Openai is designed to conduct in -depth research on a subject that is clearly quoted. This service is only available for $ 200 PRO of CHATGPT. Openai recommends everything from science to shopping research, but be careful that hallucinations are a problem for AI.
Cat Mistral
Mistral has released the App version of Le Chat, a multimodal AI personal assistant. Mistral claims that Le Chat responds faster than other chatbots. There is also a paid version with the latest journalism of AFP. Le Monde’s test is impressive, but there was more error than CHATGPT.
Openai operator
Openai’s operator is an individual intern that can work independently as it helps to buy groceries. You need $ 200 a month. Chatgpt Pro subscription is required. AI agents have many promises, but are still experimental. The Washington Post Reviewer says the operator has decided to order 12 eggs for $ 31 paid as a credit card for the reviewer.
Google Gemini 2.0 Pro Experiment
Google Gemini’s long -awaited flagship model says it’s excellent for coding and understanding common knowledge. It also has a very long context window of 2 million tokens to help users who need to quickly handle large -scale text masses. This service requires a $ 19.99 Google One Ai Premium subscription a month.
The AI model was released in 2024
DeepSeek R1
The Chinese AI model took the silicon valley to the storm. DeepSeek’s R1 means that it is well performed in coding and mathematics, while open source characteristics can be run local. Besides, it’s free. However, the R1 is rising to integrate the Chinese government censorship and potentially return user data to China.
Twins deep studies
Deep Research summarizes Google’s search results in simple and well -quoted documents. This service helps students and others who need rapid research summary. But the quality is not as good as the actual colleague review paper. Deep Research requires $ 19.99 Google One Ai Premium subscription.
The goal is 3.3 70b
This is the latest version of META’s open source LLAMA AI model. The meta is the cheapest and most efficient, but especially as mathematics, general knowledge and the following. Free and open source.
Openai sora
SORA is a model that creates a realistic video based on text. Openai can generate full scenes, not clips, but Openai often admits that it often produces “unrealistic physics.” Starting from plus, which is currently $ 20 per month, it can only be used in the paid version of CHATGPT.
Alibaba QWEN QWQ-32B-PrreView
This model is one of the few people for competing with OPENAI’s O1 on a specific industry benchmark and excellent mathematics and coding. Alibaba ironically says that in the case of “reasoning model”, it is “room for improvement of common sense reasoning.” It also includes Chinese government censorship and Techcrunch test shows. Free and open source.
Use of computers in Anthropic
Claude’s computer use is controlled by the computer to complete the tasks such as airplane ticket coding or reservation to become the predecessor of the Openai operator. But computer use remains in beta. The price is through the API: $ 0.80 per million dollars and $ 4 per 1 million dollar tokens.
X.AI’s Grok 2
X.AI, an AI company of ELON Musk, has launched an improved version of the “three times faster” flagship GrOK 2 CHATBOT. Free users are limited to 10 questions every two hours in Grok, and X’s premium and premium+ planned subscribers enjoy higher use restrictions. X.AI also launched Aurora, an image generator, and created a very photographic image, including some graphics or violent content.
Openai O1
The O1 family of Openai is to create a better answer by “thinking” through response through hidden reasoning functions. This model is excellent for coding, mathematics and safety, but Openai claims are also a problem of deceiving humans. To use O1, you must join the Chatgpt Plus, which is $ 20 a month.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is the best in the class. It becomes famous for its coding function and is considered a chatbot of an insiders. Heavy users need to join $ 20 per month, but they can access this free model in Claude. You can understand the image, but you can’t create an image.
Openai GPT 4o-Mini
Openai played GPT 4O-Mini as the cheapest and fastest model, but it was promoted thanks to the small size. It enables a wide range of work, such as supplying power to customer service chatbots. This model is available in the free tier of CHATGPT. It is more suitable for a large amount of simple tasks compared to more complex tasks.
Cosher command R+
Cohere’s Command R+ MODEL is excellent in complex search generation generation (or RAG) applications for companies. In other words, you can find and quote certain information well. (The inventor of the RAG actually works in Cohere.) RAG still does not completely solve the hallucinations of AI.
