New and recently funded AI Startups
1
Country: USA | Funding: $337M
Odyssey is an AI lab focused on creating universal world models, which the company calls new form of audiovisual intelligence. These models are to form the basis of the next generation of games, films, education content, training simulations and advertising. These models don't just generate video - they enable interaction with the 3D world i.e. creation of interactive videos. The company achieves this through a new multi-stage training pipeline that transforms the model into a causal behavioral video model that reacts to actions in real time and continuously responds to input. As you play the video, you shape it in real time using natural text prompts, similar to communicating with a language model.
Odyssey is an AI lab focused on creating universal world models, which the company calls new form of audiovisual intelligence. These models are to form the basis of the next generation of games, films, education content, training simulations and advertising. These models don't just generate video - they enable interaction with the 3D world i.e. creation of interactive videos. The company achieves this through a new multi-stage training pipeline that transforms the model into a causal behavioral video model that reacts to actions in real time and continuously responds to input. As you play the video, you shape it in real time using natural text prompts, similar to communicating with a language model.
2
Country: USA | Funding: $501.7M
Rubicon provides waste management platform and services for businesses and smart cities. Using computer vision, the platform conducts an initial analysis of a client's waste stream to identify waste types and volumes, then creates a customized waste collection schedule and optimal routes for fleets. Rubicon uses data analysis and AI to improve waste management strategies, providing transparency into waste generation patterns and predicting future trends. The company's services cover the recycling of cardboard, plastic, paper, metal, glass, electronics, construction and demolition waste, organic waste (including food waste and composting services), and separate waste collection. Acquired by Dev.al
Rubicon provides waste management platform and services for businesses and smart cities. Using computer vision, the platform conducts an initial analysis of a client's waste stream to identify waste types and volumes, then creates a customized waste collection schedule and optimal routes for fleets. Rubicon uses data analysis and AI to improve waste management strategies, providing transparency into waste generation patterns and predicting future trends. The company's services cover the recycling of cardboard, plastic, paper, metal, glass, electronics, construction and demolition waste, organic waste (including food waste and composting services), and separate waste collection. Acquired by Dev.al
3
Country: China | Funding: $7,4B
DeepSeek is specializing in the development of large-scale open-source language models. The company unexpectedly grew out of a Chinese hedge fund and disrupted the AI market with its technologies, which significantly reduced the cost of training and inference of LLM models. These technologies include Mixture of Experts - a fragmented model splitting into individual "experts" so that only a small portion of the model could be used for each task. Unlike traditional language models, which rely heavily on supervised fine-tuning, DeepSeek relies primarily on reinforcement learning. It also came up with the idea of using distilled data from OpenAI models for training. DeepSeek provides its online chatbot and an API for business. Its responses are often criticized due to Chinese censorship.
DeepSeek is specializing in the development of large-scale open-source language models. The company unexpectedly grew out of a Chinese hedge fund and disrupted the AI market with its technologies, which significantly reduced the cost of training and inference of LLM models. These technologies include Mixture of Experts - a fragmented model splitting into individual "experts" so that only a small portion of the model could be used for each task. Unlike traditional language models, which rely heavily on supervised fine-tuning, DeepSeek relies primarily on reinforcement learning. It also came up with the idea of using distilled data from OpenAI models for training. DeepSeek provides its online chatbot and an API for business. Its responses are often criticized due to Chinese censorship.
4
Country: USA | Funding: $3.4B
Anysphere was the first company to create an AI-based IDE for software development - Cursor. Developers can use AI-powered auto-completion (for both individual strings and entire classes) and ask side-chat to create or fix a specific function, class or entire file (what later became known as vibe-coding). They can also create AI agents that independently perform simple coding tasks. Cursor relies on external AI models from companies including Google, OpenAI and Anthropic to power its platform, but plans to gradually implement its own model. Developers can use different LLMs for different tasks and pay only for Cursor (there's no need to purchase a subscription for each LLM). Acquired by SpaceX
Anysphere was the first company to create an AI-based IDE for software development - Cursor. Developers can use AI-powered auto-completion (for both individual strings and entire classes) and ask side-chat to create or fix a specific function, class or entire file (what later became known as vibe-coding). They can also create AI agents that independently perform simple coding tasks. Cursor relies on external AI models from companies including Google, OpenAI and Anthropic to power its platform, but plans to gradually implement its own model. Developers can use different LLMs for different tasks and pay only for Cursor (there's no need to purchase a subscription for each LLM). Acquired by SpaceX
5
Country: USA | Funding: $9M
Probably develops technology that can prevent LLM hallucinations and simple factual errors. It's a sophisticated security system, which the startup describes as a "mechanical data suit." LLM results are verified using a deterministic validation system that returns any results that don't match the dataset. Each result is accompanied by a link to the source and an audit trail of the development process, a practice that is becoming increasingly common among AI tools.
Probably develops technology that can prevent LLM hallucinations and simple factual errors. It's a sophisticated security system, which the startup describes as a "mechanical data suit." LLM results are verified using a deterministic validation system that returns any results that don't match the dataset. Each result is accompanied by a link to the source and an audit trail of the development process, a practice that is becoming increasingly common among AI tools.
6
Country: Malaysia | Funding: $71.3M
Respond develops AI agent-powered messaging app with its customer conversation management features. The platform helps mid-to-large B2C businesses automate customer conversations across multiple messaging channels including WhatsApp, Instagram, TikTok, Messenger, Line, Telegram, WeChat, voice calls and web chat. It uses AI agents to automatically handle high volumes of customer inquiries, qualify leads and close sales without human intervention. Unlike enterprise software competitors that charge per seat, Respond charges based on the volume of customer conversations, so it doesn’t matter whether a human or an AI is answering. Respond currently more popular in APAC and Latin America countries.
Respond develops AI agent-powered messaging app with its customer conversation management features. The platform helps mid-to-large B2C businesses automate customer conversations across multiple messaging channels including WhatsApp, Instagram, TikTok, Messenger, Line, Telegram, WeChat, voice calls and web chat. It uses AI agents to automatically handle high volumes of customer inquiries, qualify leads and close sales without human intervention. Unlike enterprise software competitors that charge per seat, Respond charges based on the volume of customer conversations, so it doesn’t matter whether a human or an AI is answering. Respond currently more popular in APAC and Latin America countries.
7
Country: USA | Funding: $744.4M
Socure is developing an predictive analytics AI platform for identity verification (of consumers, partners, and employees), fraud prevention, compliance, age verification, and AI-powered personnel screening. It enables the detection of complex fraud cases using artificially created and fake identities, mitigation of non-payments, risk prediction, verification of email, phone, and physical address ownership, and the transformation of complex identity relationships into actionable insights.
Socure is developing an predictive analytics AI platform for identity verification (of consumers, partners, and employees), fraud prevention, compliance, age verification, and AI-powered personnel screening. It enables the detection of complex fraud cases using artificially created and fake identities, mitigation of non-payments, risk prediction, verification of email, phone, and physical address ownership, and the transformation of complex identity relationships into actionable insights.
8
Country: USA | Funding: $2B
Thinking Machines Lab is a startup by OpenAI co-founder Mira Muratti. It develops so-called "interaction models," which the company describes as a fundamentally different type of AI interface. Instead of the step-by-step prompt-response dynamics that define most modern AI products, it processes continuous streams of audio, text, and video at 200-millisecond intervals. The idea is that they can capture the nuances of human communication - interruptions, mid-thought corrections, even pauses for reflection - in near real-time. For now, the startup maintains "a high level of security to prevent unauthorized use of its models."
Thinking Machines Lab is a startup by OpenAI co-founder Mira Muratti. It develops so-called "interaction models," which the company describes as a fundamentally different type of AI interface. Instead of the step-by-step prompt-response dynamics that define most modern AI products, it processes continuous streams of audio, text, and video at 200-millisecond intervals. The idea is that they can capture the nuances of human communication - interruptions, mid-thought corrections, even pauses for reflection - in near real-time. For now, the startup maintains "a high level of security to prevent unauthorized use of its models."
9
Country: USA
UXMagic.ai turns prompts, sketches, screenshots, Figma files, and URLs into editable UI designs, wireframes, and React code.
UXMagic.ai turns prompts, sketches, screenshots, Figma files, and URLs into editable UI designs, wireframes, and React code.
10
Country: USA | Funding: $618.5M
Lemonade uses AI to create the perfect policies for homeowners and renters and pay them insurance claims. Jim, an insurance claims processing company, processes approximately one-third of Lemonade's insurance claims autonomously. Jim uses natural language processing and machine learning algorithms to analyze insurance claims. It even makes payout decisions in seconds, significantly reducing the time and resources required to process insurance claims. The average claim settlement time was just 3 seconds. This rapid processing is due to Jim's ability to analyze claims data, evaluate insurance coverage, and make payout decisions with a high degree of accuracy.
Lemonade uses AI to create the perfect policies for homeowners and renters and pay them insurance claims. Jim, an insurance claims processing company, processes approximately one-third of Lemonade's insurance claims autonomously. Jim uses natural language processing and machine learning algorithms to analyze insurance claims. It even makes payout decisions in seconds, significantly reducing the time and resources required to process insurance claims. The average claim settlement time was just 3 seconds. This rapid processing is due to Jim's ability to analyze claims data, evaluate insurance coverage, and make payout decisions with a high degree of accuracy.
