AI models are turning at a sugar pace, from everyone from large technology companies such as Google to newly established companies such as Openai and Anthropic. Monitoring the latter can be overwhelming.
The addition to the confusion is that AI models are often promoted on the basis of industry reference points. But these technical measurements often reveal little about how real people and companies really use.
To cut the noise, TechCrunch has compiled an overview of the most advanced AI models released by 2024, with details of how to use them and what is best. We will maintain this list up -to -date with the latest starts.
There are literally over a million AI models out there: Huggingface, for example, hosts over 1.4 million. So this list may lose some models that perform best, in one way or another.
AI models released in 2025
OPENAI O3-MINI
This is OpenAi’s latest logic model and is optimized for STEM -related work such as coding, mathematics and science. Is not the most powerful model of Openai, but because it is smaller, the company she says Is significantly lower costs. Available for free, but requires a subscription for heavy users.
Openai Deep Research
Openai’s deep research is designed to do in -depth research on a topic of clear references. This service is only available with the $ 200 subscription per month. Open it recommends it For everything, from science to market research, but be careful that illusions remain a problem for AI.
Mistral le chat
Mistral launched the applications of Le Chat, a multimodal AI personal assistant. Incorrect allegations Le Chat responds faster than any other chatbot. It also has a paid version with updated journalism from AFP. Tests by Le Monde He found the performance of Le chat impressive, though he made more errors than chatgpt.
Opener
The Openai operator is meant to be a personal trainee who can do things independently, such as helping you buy groceries. It requires a $ 200 subscription a month a month Chatgpt Pro. AI agents have a lot of promises, but they are still experimental: a Washington Post critic says the exploitation carrier He decided on her own to order twelve eggs for $ 31, paid with the critic’s credit card.
Google Gemini 2.0 Pro Experimental
Google Google’s long -awaited flagship model reports that it is superior to coding and understanding general knowledge. It also has a super-monkey environment of 2 million brands, helping users who need to quickly process massive text pieces. The service requires (at least) a Google One Ai Premium subscription of $ 19.99 per month.
AI models released in 2024
Deepseek r1
This Chinese model AI took Silicon Valley from the storm. Deepseek R1 attributes well to coding and mathematics, while the nature of open source means that anyone can run it locally. In addition, it’s free. However, the R1 incorporates the censorship of the Chinese government and faces increasing prohibitions for the possible sending of user data back to China.
Gemini Deep Research
Deep research summarizes Google’s search results in a simple and well -mentioned document. The service It is useful for students and anyone else who needs a quick research summary. However, its quality is not as good as a real paper assessed by peer. Deep research requires a Google One Ai Premium Google One subscription.
Meta llama 3.3 7b
This is the newest and most advanced version of Meta’s Llama AI models. Meta has added this edition As cheaper and more effective, especially for mathematics, general knowledge and teaching. It is free and open source.
Openai sora
Sora is a model that creates realistic text -based videos. While it can create whole scenes and not just clips, Openai admits that often creates “unrealistic physics”. It is currently only available in paid versions of Chatgpt, starting with Plus which is $ 20 a month.
Alibaba qwen qwq-32b-preview
This model is one of the few that competes with O1 of O1 at some industry’s reference points, exceeding mathematics and coding. Ironically for a ‘model of reasoning’, it has ‘margin of improvement in common sense’, Says alibaba. It also incorporates the censorship of the Chinese government, the TechCrunch test shows. It is free and open source.
The use of ANTHROPIC PC
The use of Claude’s computer is intended to take control of your computer to complete tasks such as coding or booking an airline ticket, making it a predecessor of the Openai operator. Use computer, however, remains in beta. The pricing is via API: $ 0.80 per million input brands and $ 4 per million brands.
x.ai’s grok 2
X.Ai, AI owned by Elon Musk, launched an improved version of the flagship Grok 2 Chatbot It allegations is “three times faster”. Free users are limited to 10 questions every two hours in Grok, while X’s subscribers enjoy higher use limits. The X.Ai also started an image generator, Aurora, that produces excellent photorealistic images, including graphic or violent content.
OPENAI O1
Openai’s O1 family is meant to produce better answers with “thinking” through answers through a hidden characteristic reasoning. The model is superior to coding, mathematics and security, OPERAI REQUIREMENTSBut it also has problems that deceive people. O1 requires recording in Chatgpt Plus, which is $ 20 a month.
Claude Sonnet 3.5 of Anthropic
Claude Sonnet 3.5 is a model of anthropogenic claims as better in class. Has been known for coding capabilities and is considered a technological confidential Chatbot of the choice. The model can be reached free of charge to Claude, although heavy users will need a monthly pro $ 20 subscription. While he can understand the images, he cannot create them.
Openai GPT 4o-mini
Openai has offered the 4o-mini GPT as its most affordable and faster model, thanks to its small size. Mean To activate a wide range of tasks such as the supply of customer service chatbots. The model is available at the free level of ChatGPT. It is best suitable for simple large volume tasks compared to more complex.
CONHERE COMMAND R+
Agreed Command r+ model It depends on complex production applications (or rag) for businesses. This means that he can find and report specific information information very well. (The inventor of the rag Actually works on cohere.) Still, RAG does not fully solve the problem of AI’s illusions.
