AI Models: Keeping Up in a Fast-Evolving Landscape
AI models are proliferating at a rapid pace, with new releases from major tech companies and innovative startups. This surge of innovation can be overwhelming, especially when promotional materials emphasize industry benchmarks that may not reflect real-world applications. To provide clarity, TechCrunch has compiled a comprehensive overview of the most advanced AI models launched since 2024, offering details on their capabilities and best use cases. This list will be continuously updated to reflect the latest developments in this dynamic field.
It’s important to note that the AI landscape is vast, with over a million models available. This overview aims to highlight the most significant advancements, but it’s possible that other models may offer superior performance in specific areas.
AI Models Released in 2025
OpenAI’s GPT 4.5 ‘Orion’
OpenAI positions Orion as its most extensive model to date, emphasizing its strong “world knowledge” and “emotional intelligence.” Despite these capabilities, Orion’s performance lags behind newer reasoning models in some benchmarks. Orion is available through OpenAI’s $200/month subscription plan.
Claude Sonnet 3.7
Anthropic describes Sonnet 3.7 as the industry’s first hybrid reasoning model, capable of producing both quick answers and in-depth analysis as needed. Users can also control how long the model spends on complex thought processes. Sonnet 3.7 is accessible to all Claude users, although heavy users will need the $20/month Pro plan.
xAI’s Grok 3
Grok 3, the latest flagship model from Elon Musk’s xAI, is designed to excel in math, science, and coding, reportedly outperforming other leading models. Access to the model requires an X Premium subscription, priced at $50 per month. Following a study that indicated Grok 2’s leaning toward the political left, Musk has pledged to make Grok more politically neutral.
OpenAI o3-mini
OpenAI has optimized o3-mini for tasks in STEM fields, such as coding, mathematics, and science. This model is smaller than OpenAI’s most powerful models which results in significantly lower costs, according to the company. It is accessible for free with a subscription necessary for frequent usage.
OpenAI Deep Research
OpenAI’s Deep Research is designed to provide extensive research on a given subject, with clear source citations and is available through ChatGPT’s $200 monthly Pro subscription. OpenAI recommends it for an array of uses, from scientific studies to shopping research however, the company reminds users that the AI’s tendency for factual inaccuracies is something to be aware of.
Mistral Le Chat
Mistral has launched app versions of Le Chat, a multimodal AI personal assistant, boasting faster response times than competing chatbots. It also has a paid version that features up-to-date journalism from Agence France-Presse. While Le Monde’s tests found Le Chat’s performance impressive, the model did make more errors than ChatGPT.
OpenAI Operator
OpenAI’s Operator is designed as a personal intern that is able to autonomously complete tasks, such as grocery shopping. It’s powered by a $200 subscription to ChatGPT Pro plan. AI agents show a lot of promise although they are still in their early stages. As a case in point: a reviewer at the Washington Post reports that Operator independently ordered a dozen eggs for $31, using the reviewer’s credit card.
Google Gemini 2.0 Pro
Google Gemini’s flagship model is designed to excel in coding and general knowledge, incorporating a lengthy context window with up to 2 million tokens. The model requires a minimum of a Google One AI Premium subscription, which is $19.99 per month.
AI Models Released in 2024
DeepSeek R1
This Chinese-based AI model has rapidly gained attention and has become very popular in Silicon Valley. DeepSeek’s R1 excels in coding and math, and its open-source nature allows it to be run locally by anyone. It’s also free to download and use. However, it is important to note that R1 integrates Chinese government censorship and faces rising bans for potential data transfer to China.
Gemini Deep Research
Google’s Deep Research service provides simple, well-cited summaries of Google search results. It’s designed for students and anyone needing quick research, but it is worth noting that the quality is below that of a traditional peer-reviewed paper. Deep Research does require the $19.99 Google One AI Premium subscription.
Meta Llama 3.3 70B
Meta’s open-source Llama AI models have been updated to include the newest version. This version is the most advanced to date, and Meta highlights its efficiency and cost-effectiveness, especially for math, general knowledge, and instruction following. This model is free and open source.
OpenAI Sora
Sora is a model developed to create realistic videos based on user-provided text prompts. While it can create entire scenes rather than just short clips, OpenAI acknowledges that it sometimes produces unrealistic outcomes. It’s currently available on paid versions of ChatGPT starting with the Plus plan, which is $20 per month.
Alibaba Qwen QwQ-32B-Preview
This model has become one of the few models to rival OpenAI’s o1 on select industry benchmarks, particularly in math and coding. Despite the fact that it is considered a ‘reasoning model,’ Alibaba reports that there is some ‘room for improvement in common sense reasoning.’ TechCrunch testing has revealed that this model currently incorporates Chinese government censorship. It is free and open source.
Anthropic’s Computer Use
Anthropic’s Computer Use is designed to take control of a computer to complete tasks, like programming or making travel arrangements, functioning similarly to OpenAI’s Operator. But, Computer Use is still in its beta phase. Pricing is determined via API: it costs $0.80 per million tokens of input and $4 per million tokens of output.
x.AI’s Grok 2
Elon Musk’s AI company, xAI, released an updated version of its flagship Grok 2 chatbot, which claims to be three times faster. Free users are limited to 10 questions every two hours on Grok, while users of X’s Premium and Premium+ plans benefit from higher usage limits. x.AI also launched an image generator, Aurora, capable of producing highly photorealistic images, some of which contain graphic content.
OpenAI o1
OpenAI’s o1 family of models is designed to provide better answers by using a hidden reasoning feature. OpenAI claims that the model excels in math, coding, and security, but it can at times be inconsistent when generating answers. Access requires a subscription to ChatGPT Plus, which is $20 per month.
Anthropic’s Claude Sonnet 3.5
Anthropic’s Claude Sonnet 3.5 is a model Anthropic claims is best in class. It is lauded for its impressive coding abilities and is considered popular among tech insiders as the chatbot of choice. The model can be accessed for free on Claude, but a $20 monthly Pro subscription is required for heavy users. Though it can interpret images, it is unable to generate them.
OpenAI GPT 4o-mini
OpenAI has stated that GPT 4o-mini is its fastest and most affordable model, thanks to its compact size. It’s capable of carrying out a variety of tasks, such as powering customer service chatbots. This model is offered on the free tier of ChatGPT. It’s more suitable for high-volume simple tasks compared to more complex ones.
Cohere Command R+
Cohere’s Command R+ model excels in complex Retrieval-Augmented Generation (or RAG) applications for enterprises. Its strength lies in its precise location and citation of specific pieces of information (Cohere employs the inventor of RAG), though it is important to know that it does not completely eliminate AI’s hallucination problem.
