AI models are being developed at an astonishing rate by a wide array of entities, from major tech corporations like Google to innovative startups such as OpenAI and Anthropic. Keeping up with these rapid advancements can feel daunting.
Complicating the situation further, AI models are frequently evaluated based on industry standards. However, these technical benchmarks often provide minimal insight into how actual users and businesses leverage these tools.
To help clarify the landscape, TechCrunch has assembled a rundown of the most cutting-edge AI models that have emerged since 2024, including guidance on their applications and strengths. This list will be continuously updated to include the latest releases.
There are actually over a million AI models currently accessible; for instance, Hugging Face boasts over 1.4 million. Consequently, this compilation may overlook certain models that could outperform others in various aspects.
AI Models Introduced in 2025
Cohere’s Aya Vision
Cohere has introduced a multimodal model named Aya Vision, which it proudly claims excels at tasks like image captioning and answering inquiries related to photos. Unlike many other models, it demonstrates proficiency in multiple languages, including those outside of English. Users can access it for free on WhatsApp.
OpenAI’s GPT 4.5 ‘Orion’
OpenAI has labeled Orion as their most extensive model yet, highlighting its impressive “world knowledge” and “emotional intelligence.” Nevertheless, it falls short on certain benchmarks when compared to newer reasoning models. Orion can be accessed by subscribers to OpenAI’s $200 monthly plan.
Claude Sonnet 3.7
Anthropic has declared this to be the industry’s inaugural ‘hybrid’ reasoning model, capable of delivering rapid answers while engaging in more thorough contemplation as necessary. It offers users the ability to determine how long the model remains engaged, as per Anthropic. Sonnet 3.7 is available for all Claude users, though frequent users will need a $20 monthly Pro subscription.
xAI’s Grok 3
The latest flagship model from xAI, founded by Elon Musk, is Grok 3. It is promoted as outperforming its competitors in areas like math, science, and coding. Access to this model requires an X Premium subscription, costing $50 per month. Following findings that Grok 2 exhibited a left-leaning bias, Musk committed to steering Grok towards a more “politically neutral” stance, although it is yet to be determined if this goal has been accomplished.
OpenAI o3-mini
OpenAI’s latest reasoning model, o3-mini, is fine-tuned for STEM-related tasks, including coding, math, and scientific analysis. This model is not the most powerful from OpenAI but, being smaller, is touted as more cost-effective. It is free to use, though a subscription is needed for frequent users.
OpenAI Deep Research
OpenAI Deep Research is tailored for conducting thorough research on various topics with reliable citations. This feature is exclusively available with ChatGPT’s $200 monthly Pro plan. OpenAI recommends it for pursuits ranging from scientific studies to shopping research, but users should remain cautious, as AI-generated content can still include inaccuracies.
Mistral Le Chat
Mistral has unveiled app versions of Le Chat, its multimodal AI personal assistant. Mistral claims that Le Chat responds more swiftly than competing chatbots. It also offers a premium version featuring real-time journalism provided by AFP. According to evaluations from Le Monde, Le Chat demonstrated impressive performance yet made more errors than ChatGPT.
OpenAI Operator
OpenAI’s Operator is designed to function as a personal intern, capable of independently handling tasks like grocery shopping. A $200 monthly subscription to ChatGPT Pro is required for access. Although AI agents show great potential, they remain experimental; a Washington Post review noted that Operator autonomously decided to purchase a dozen eggs for $31 using the reviewer’s credit card.
Google Gemini 2.0 Pro Experimental
The eagerly anticipated flagship model from Google Gemini boasts exceptional performance in coding and general knowledge comprehension. It features an extensive context window of 2 million tokens, assisting users with the rapid processing of large quantities of text. A minimum subscription of $19.99 to Google One AI Premium is required for access.
AI Models Released in 2024
DeepSeek R1
This Chinese AI model has made significant waves in Silicon Valley. DeepSeek’s R1 excels in coding and mathematics, with its open source format allowing users to run it on local machines for free. However, it integrates with Chinese government censorship and is facing increased scrutiny for potentially transmitting user data back to China.
Gemini Deep Research
Gemini Deep Research compiles Google’s search results into a straightforward, well-cited document. This service is beneficial for students and anyone else needing a concise research summary. However, its quality does not match that of actual peer-reviewed research papers. Gemini Deep Research necessitates a $19.99 Google One AI Premium subscription.
Meta Llama 3.3 70B
This is the most recent and advanced iteration of Meta’s open-source Llama AI models. Meta has described this version as both cost-effective and efficient, particularly for mathematical queries and task-following instructions. It is free to use and open source.
OpenAI Sora
Sora is a model designed to generate realistic videos based on text prompts. While it is capable of creating complete scenes instead of merely clips, OpenAI acknowledges that it frequently produces “unrealistic physics.” Currently, it is only available with paid subscriptions of ChatGPT, starting at $20 per month with the Plus plan.
Alibaba Qwen QwQ-32B-Preview
This model gestures towards being a competitor to OpenAI’s o1, especially in specialized industry benchmarks where it shines in math and coding. Ironically, this “reasoning model” has “opportunities for improvement in commonsense reasoning,” claims Alibaba. Additionally, it incorporates Chinese governmental censorship as indicated by TechCrunch testing. This model is also free and open source.
Anthropic’s Computer Use
Claude’s Computer Use aims to manage your computer autonomously for tasks such as coding or booking flights, acting as a precursor to OpenAI’s Operator. This feature still exists in beta. Pricing is set via API at $0.80 per million tokens input and $4 per million tokens output.
x.AI’s Grok 2
Elon Musk’s x.AI has released an enhanced version of its Grok 2 chatbot, which it asserts is “three times faster.” Free users can ask up to 10 questions every two hours on Grok, while subscribers of X’s Premium and Premium+ tiers have access to higher question limits. Additionally, x.AI has rolled out an image generation tool, Aurora, capable of producing highly realistic images, including some potentially graphic or violent content.
OpenAI o1
OpenAI’s o1 series is geared towards delivering improved responses by employing a hidden reasoning feature. The model is especially adept at coding, mathematics, and maintaining safety, as claimed by OpenAI. Using o1 requires a subscription to ChatGPT Plus, priced at $20 monthly.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is promoted by Anthropic as a leading model, recognized for its coding capabilities and has become a preferred choice among tech insiders for chatbot interactions. While it allows free access on Claude, frequent users will need to opt for a $20 monthly Pro subscription. The model can interpret images but does not have the ability to create them.
OpenAI GPT 4o-mini
OpenAI describes GPT 4o-mini as their most cost-effective and quickest model to date, owing to its compact size. Intended to facilitate a variety of tasks including powering customer service chatbots, this model is available on ChatGPT’s free tier. It is more suited for high-volume, straightforward tasks rather than complex operations.
Cohere Command R+
Cohere’s Command R+ model shines in complex Retrieval-Augmented Generation (RAG) applications targeted at enterprises, allowing it to efficiently locate and cite specific information. (The pioneer of RAG currently works at Cohere.) That said, RAG does not completely eliminate the hallucination challenges faced by AI.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
