AI models are being developed at an astonishing rate by a wide array of entities, ranging from major technology corporations such as Google to innovative startups like OpenAI and Anthropic. Keeping pace with the most recent advancements can be quite daunting.
The situation becomes even more complicated as AI models are often marketed based on specific industry standards. Yet, these technical specifications frequently provide minimal insight into how actual users and businesses interact with these technologies.
To simplify matters, TechCrunch has gathered a summary of the most cutting-edge AI models that have been launched since 2024, complete with practical usage tips and their main strengths. We will continually update this list to include the newest releases.
There are currently over a million AI models available; for instance, Hugging Face houses more than 1.4 million. Thus, it is possible that this compilation may overlook some models that could perform better in various contexts.
AI Models Released in 2025
OpenAI’s GPT 4.5 ‘Orion’
OpenAI describes Orion as their most expansive model yet, boasting impressive “world knowledge” and “emotional intelligence.” However, it does not score as well on certain performance metrics compared to more recent logical models. Orion is accessible to users on OpenAI’s $200 monthly subscription plan.
Claude Sonnet 3.7
Anthropic claims this model is the first ‘hybrid’ reasoning model in the industry, capable of delivering rapid responses and engaging in deeper analytic thinking when necessary. Users can customize the duration for which the model can analyze, according to Anthropic. Sonnet 3.7 is available to all Claude users, though frequent users will need to subscribe to a $20 monthly Pro plan.
xAI’s Grok 3
Grok 3 is the flagship model from xAI, the AI firm founded by Elon Musk. It is reported to surpass other leading models in tasks involving mathematics, science, and programming. Access requires a subscription to X Premium at $50 per month. After a study indicated that Grok 2 had a left-leaning bias, Musk promised to make Grok more “politically neutral,” though it remains unclear if this goal has been met.
OpenAI o3-mini
This is OpenAI’s newest reasoning model, designed with a focus on STEM tasks such as coding, math, and science. It is not the most powerful model from OpenAI, but its smaller size allows for significantly reduced costs, making it available for free, albeit requiring a subscription for heavy users.
OpenAI Deep Research
Deep Research from OpenAI is crafted for conducting comprehensive research on various topics with well-defined citations. This feature is only available through ChatGPT’s $200 monthly Pro subscription. OpenAI suggests using it for purposes ranging from scientific inquiry to product research, but users should remain cautious as AI hallucinations still pose challenges.
Mistral Le Chat
Mistral has rolled out app versions of Le Chat, a versatile multimodal AI assistant. Mistral claims that Le Chat offers faster responses than competing chatbots. A premium version includes current journalism sourced from AFP. According to tests by Le Monde, Le Chat demonstrates notable performance, although it made more mistakes than ChatGPT.
OpenAI Operator
OpenAI’s Operator is designed to act as a personal assistant capable of executing tasks independently, such as grocery shopping. A ChatGPT Pro subscription at $200 per month is required for access. While AI agents show significant potential, they are still in experimental phases; a reviewer from the Washington Post noted that Operator once autonomously ordered a dozen eggs for $31, using the reviewer’s credit card.
Google Gemini 2.0 Pro Experimental
Google Gemini’s much-anticipated premier model claims to excel in coding and general knowledge comprehension. With an impressively extended context window of 2 million tokens, it aids users needing to manage vast amounts of information quickly. This service requires a minimum of a Google One AI Premium subscription, priced at $19.99 per month.
AI Models Released in 2024
DeepSeek R1
This Chinese AI model has made waves in Silicon Valley. DeepSeek’s R1 excels in math and programming, while its open-source nature allows users to run it locally without charge. However, it is subject to Chinese government censorship and faces increasing restrictions due to concerns about potential data leaks to China.
Gemini Deep Research
Gemini Deep Research synthesizes Google’s search results into easily understandable, well-cited documents. This tool is particularly beneficial for students and others in need of quick research synopses, though its output does not match the quality of peer-reviewed academic work. Gemini Deep Research also requires a $19.99 monthly subscription for Google One AI Premium.
Meta Llama 3.3 70B
This represents the latest iteration of Meta’s open-source Llama AI models. Meta has marketed this version as its most economical and efficient to date, excelling in areas like math, general knowledge, and following instructions. It is available free of charge as open-source content.
OpenAI Sora
Sora is a model that produces realistic video content from text prompts. While capable of creating entire scenes instead of just clips, OpenAI has acknowledged that it frequently generates “unrealistic physics.” This model is accessible only through the paid tiers of ChatGPT, starting with Plus at $20 a month.
Alibaba Qwen QwQ-32B-Preview
This model is among the few that directly compete with OpenAI’s o1 in specific industry benchmarks, showcasing strengths in math and programming. Notably, despite being branded as a “reasoning model,” it has “room for improvement in common sense reasoning,” according to Alibaba. It also incorporates features of Chinese government censorship, as shown by testing conducted by TechCrunch. This model is also available free of charge and as open source.
Anthropic’s Computer Use
Claude’s Computer Use is designed to perform tasks on your computer, such as coding or booking flights, serving as a precursor to OpenAI’s Operator. However, Computer Use remains in beta. Pricing is based on an API model: $0.80 per million tokens of input and $4 per million tokens of output.
x.AI’s Grok 2
Elon Musk’s x.AI recently unveiled a refined version of its flagship Grok 2 chatbot, which it claims is “three times faster.” Free users are limited to asking 10 questions every two hours, while subscribers of X’s Premium and Premium+ plans enjoy more expansive usage allowances. Additionally, x.AI launched Aurora, an image generator that creates highly lifelike images, which may include graphic or violent content.
OpenAI o1
The o1 family from OpenAI is crafted to yield superior answers by employing a hidden reasoning capability. This model is said to excel in coding, math, and safety, as noted by OpenAI, but it also faces challenges in deceiving humans. Access to o1 necessitates a subscription to ChatGPT Plus, costing $20 per month.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is presented by Anthropic as a top-tier model, known for its strong coding skills and considered a go-to chatbot in the tech industry. The model is available for free on Claude, although more intensive users will need a $20 monthly Pro subscription. While it can comprehend images, it does not have the capability to generate them.
OpenAI GPT 4o-mini
OpenAI has promoted GPT 4o-mini as its most cost-effective and rapid model to date, thanks to its compact structure. It is designed to facilitate a wide array of functions, including powering customer service chatbots. This model can be accessed through ChatGPT’s free tier, being better suited for high-volume, straightforward tasks rather than complex ones.
Cohere Command R+
Cohere’s Command R+ model is adept at complex Retrieval-Augmented Generation (RAG) applications within enterprises. This means it excels at finding and accurately citing specific pieces of information. (Interestingly, the inventor of RAG currently works at Cohere.) Nevertheless, RAG does not entirely rectify AI’s hallucination issue.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
