General Intuition Review: Key Features and Pros&Cons

Name: General Intuition
Availability: InStock
Author: General Intuition

What it is:General Intuition is a frontier AI research lab spun out from Medal, building foundation models and agents for spatial-temporal reasoning in video games using billions of gameplay clips.
Best for:AAA game studios, Competitive multiplayer developers, AI research teams in robotics/gaming
Rating:85/100Very Good
Expert's conclusion:An excellent research lab for developing innovative embodied AI technology; however, a viable product for either a commercial or development entity today it is not.

Reviewed byMaxim Manylov·Web3 Engineer & Serial Founder

Company Overview

General Intuition is a frontier AI research laboratory that was established by Medal.tv. It is an AI laboratory developing foundation models with deep spatial and temporal reasoning abilities for both gaming AI and other uses. The AI laboratory in New York utilizes billions of clips of games for training AI agents that can see and predict behaviors in many different environments solely based upon visual information. In addition to enhancing gaming experiences using AI powered NPC's and simulation tools, General Intuition also has a focus on utilizing its technology to develop applications for use in robotics and autonomous systems.

Active

📍New York City, NY

📅Founded 2025

🏢Private

TARGET SEGMENTS

Gaming IndustryAI ResearchRoboticsAutonomous Vehicles

Key Metrics

📊

$133.7M

Seed Funding

📊

2B+ per year

Data Clips

👥

10M

Monthly Active Users (Medal)

📊

H1 2026

Commercial Deployment Target

Credibility Rating

85/100

Excellent

Early rounds of funding from tier one venture capitalists (VCs) combined with a solid technical base developed through utilization of a massive proprietary dataset have positioned General Intuition as a highly promising participant in the field of gaming AI and spatial reasoning.

BREAKDOWN

Product Maturity75/100

Company Stability95/100

Security & Compliance70/100

User Reviews60/100

Transparency85/100

Support Quality70/100

TRUST SIGNALS

$133.7M seed funding from Khosla Ventures and General CatalystSpun out from Medal.tv with 2B+ annual gameplay clips datasetTeam includes world modeling researchers (DIAMOND, Δ-IRIS, IRIS)Targeted H1 2026 commercial deployment of AI NPCsPublic-benefit corporation committed to enhancing (not replacing) game developers

Company History

2025

Company Founded

General Intuition was established by spinning out as a frontier AI research laboratory from Medal.tv; the primary focus of the organization is developing AI technologies specifically for gaming and spatial/temporal reasoning.

2025

$133.7M Seed Funding

General Intuition raised a large seed round of funding which was led by Khosla Ventures and General Catalyst, and included the participation of The Raine Group --one of the largest investments in AI for 2025.

2025

OpenAI Acquisition Interest

According to reports, OpenAI attempted to purchase Medal.tv for approximately $500 million, indicating the strategic importance of the data assets utilized by the company for training its AI agents.

2026

Commercial Launch Target

By the end of H1 2026, General Intuition plans to deploy AI-powered non-player characters (NPCs) and interactive simulation tools into the market.

Key Executives

Pim de Witte— CEO & Co-founder: Founder of Medal.tv and a Dutch entrepreneur who worked in humanitarian relief prior to creating Medal.tv. The experience working in humanitarian relief has informed General Intuition's focus on search and rescue applications in addition to its focus on gaming AI.
Moritz Baier-Lentz— Founding Member: A partner at Lightspeed Ventures with extensive knowledge of gaming AI. He is interested in scaling AI bots that are able to adapt to a player's level of ability to optimize their engagement with the game.

Key Features

✨

Spatial-Temporal Reasoning

The foundation models developed by General Intuition were trained on over two billion clips of gameplay, enabling them to understand how objects move and interact with each other within environments through the visual input only.

✨

Visual-First AI Agents

The agents used by General Intuition perceive their environment exactly like a human does when they view their surroundings through a first-person camera view, and can predict the behavior of their environment and navigate unfamiliar areas.

📊

Adaptive Gaming NPCs

The AI-powered NPCs used by General Intuition will scale dynamically to a player's level of ability, and maintain a ~50% win rate in order to maximize player engagement.

✨

Cross-Domain Transfer

Gaming-trained models are applied to physical systems such as robots, drones, and self-driving cars using a controller-based paradigm for navigation.

✨

Interactive World Simulation

Simulated environments are created for agents to be trained in while avoiding copyright problems by emphasizing agent behavior rather than content creation.

✨

Edge Case Training

Highlight clips from gamers are used to create selection-biased extreme examples of what an AI can learn from, which are ideal for robust AI training.

Tech Stack

Infrastructure

Multi-office setup (New York City, Geneva); research-focused infrastructure

Technologies

PyTorchReinforcement LearningComputer Vision

Integrations

Gameplay Data PlatformsSimulation EnvironmentsRobotics Hardware

AI/ML Capabilities

Proprietary foundation models specializing in spatial-temporal reasoning, world modeling (DIAMOND, Δ-IRIS, IRIS influences), and video understanding trained exclusively on visual gameplay data for agentic systems.

Inferred from research publications mentioned and gaming AI focus; specific stack not publicly detailed

Use Cases

Game Developers

Adaptive AI NPC’s that can adapt to the skill of any player are deployed to enhance retention while providing an opponent that is engaging regardless of the game scenario.

Game Studios

Tools are developed to test game mechanics and environments using AI generated worlds that have been trained with data from actual gameplay.

Search-and-Rescue Teams

The spatial-temporal reasoning trained from gaming data is applied to power GPS-denied drones that navigate unfamiliar areas using visual-spatial reasoning.

Autonomous Vehicle Researchers

A controller-based paradigm for manipulating objects physically is utilized to transfer spatial-temporal reasoning trained from gaming data to physical navigation systems.

Robotics Engineers

Visual-only agent training is applied to robotic arms and manipulators operating in unstructured environments.

NOT FORContent Creators

Not Recommended - Company explicitly does not license its world model to prevent copyright issues when generating content in generative content applications.

NOT FORGame Asset Generators

Not Suitable - Focuses on agent behavior and reasoning to train agents, rather than developing the game assets, characters, and environments.

Pricing

Pricing information with service tiers, costs, and details
☐Service	$Cost	ℹDetails	🔗Source
Commercial Access		Pre-launch product focused on AI research. Public release expected H1 2026 with pricing TBD.	Company announcements and TechCrunch
Research Partnerships	Custom enterprise agreements	B2B licensing for AI agents, NPCs, and spatial reasoning models. Contact sales for quotes.	—

Commercial Access

Pre-launch product focused on AI research. Public release expected H1 2026 with pricing TBD.

Company announcements and TechCrunch

Research PartnershipsCustom enterprise agreements

B2B licensing for AI agents, NPCs, and spatial reasoning models. Contact sales for quotes.

Competitive Comparison

Feature	General Intuition	DeepMind Genie	World Labs Marble	OpenAI Sora
Core Functionality	Spatial-temporal reasoning via game data	World model generation	Interactive world simulation	Video generation
Training Data	Video game footage (Medal.tv)	Synthetic video data	Real-world video synthesis	Web-scale video
Primary Use Case	AI agents/NPCs for games	Agent training & content	Simulation & content	Video generation
Pricing	Not available (pre-launch)	Research access only	Enterprise licensing	$200+/month API
Free Tier	No	No	No	No
Enterprise Features	Custom training pipelines	—	Custom models	SSO, audit logs
API Availability	H1 2026 planned	Research API	Enterprise API	Yes
Integration Count	Gaming ecosystems	ML frameworks	Unity/Unreal	Developer tools
Support Options	Enterprise contracts	Research community	Enterprise	Priority support
Security Certifications	—	Google Cloud	—	SOC 2

Core Functionality

General IntuitionSpatial-temporal reasoning via game data

DeepMind GenieWorld model generation

World Labs MarbleInteractive world simulation

OpenAI SoraVideo generation

Training Data

General IntuitionVideo game footage (Medal.tv)

DeepMind GenieSynthetic video data

World Labs MarbleReal-world video synthesis

OpenAI SoraWeb-scale video

Primary Use Case

General IntuitionAI agents/NPCs for games

DeepMind GenieAgent training & content

World Labs MarbleSimulation & content

OpenAI SoraVideo generation

Pricing

General IntuitionNot available (pre-launch)

DeepMind GenieResearch access only

World Labs MarbleEnterprise licensing

OpenAI Sora$200+/month API

Free Tier

General IntuitionNo

DeepMind GenieNo

World Labs MarbleNo

OpenAI SoraNo

Enterprise Features

General IntuitionCustom training pipelines

DeepMind Genie—

World Labs MarbleCustom models

OpenAI SoraSSO, audit logs

API Availability

General IntuitionH1 2026 planned

DeepMind GenieResearch API

World Labs MarbleEnterprise API

OpenAI SoraYes

Integration Count

General IntuitionGaming ecosystems

DeepMind GenieML frameworks

World Labs MarbleUnity/Unreal

OpenAI SoraDeveloper tools

Support Options

General IntuitionEnterprise contracts

DeepMind GenieResearch community

World Labs MarbleEnterprise

OpenAI SoraPriority support

Security Certifications

General Intuition—

DeepMind GenieGoogle Cloud

World Labs Marble—

OpenAI SoraSOC 2

Competitive Position

vs DeepMind Genie

While both General Intuition and DeepMind develop world models, they differ in their approach with General Intuition focusing on the use of real gameplay data to emphasize temporal reasoning in game-native spatial reasoning versus World Labs' focus on interactive simulations; General Intuition was spun out of Medal.tv giving it a data advantage in the area of gaming but a much narrower scope.

The GI Lab has a greater emphasis on developing new ways of using machine learning to create intelligent game agents than does World Labs.

vs World Labs Marble

While both General Intuition and World Labs develop world models, General Intuition focuses on the temporal reasoning in games whereas World Labs focuses on the interactive nature of simulation; General Intuition has a data moat related to gaming due to being spun from Medal.tv but a much narrower focus than World Labs.

World Labs provides a creative environment that allows developers to simulate worlds and try out different ideas. It's well suited to tasks such as level design and world-building.

vs OpenAI Sora

Sora excels in video generation but lacks agent training focus; General Intuition focuses on agent training/agent reasoning/skills related to spatial reasoning and generation and not video generation; OpenAI has a broader ecosystem in the area of AI development and deployment but General Intuition is focused on games and therefore has a much narrower scope.

Sora is an AI platform designed specifically for generating media content such as images, videos and text whereas GI is primarily used to develop intelligent game agents.

vs NVIDIA ACE

GI is more focused on developing future reasoning technologies than providing commercialized solutions like NVIDIA's.

NVIDIA provides the technology required for producing high quality games. The technology developed by GI will enable the development of intelligent game agents with the ability to reason about their environment in a way that humans do.

Pros Cons

Pros

GI raised over $134 million in a single round of seed funding which makes it one of the most heavily backed companies in the spatial AI space.
Medal.tv holds the exclusive rights to the largest collection of gameplay footage in the world with over millions of hours of gameplay available to be analyzed by its spatial reasoning AI.
Game native training approaches allow the use of a company's own proprietary game data to train the AI rather than relying on generic data that may not be applicable to the specific game being developed.
A scalable difficulty AI opponent means that the AI can be adjusted to play against all skill levels. For example, if a game has a large number of novice players then the AI could be made easier so that these players have an opportunity to compete and stay engaged.
GI was spun off from Medal.tv which has a proven track record of processing large amounts of video at scale.
GI has attracted some of the biggest names in venture capital including Khosla and General Catalyst. This is an indication that there are big plans for the company and that the backers believe it has huge potential.
However, GI has stated that they will not make their product commercially available until H1 2026. Until this time, the product is only available internally for testing purposes.

Cons

As GI currently focuses solely on the gaming market, it is likely that there will be limited application for the products and services offered by GI outside of gaming.
Unfortunately, there is currently no publicly available pricing information for GI's products and services. As GI will only offer their products to enterprise customers, the cost of utilizing their products and services is completely unknown.
At present, GI is still in the early stages of research and development. Therefore, it is yet to be seen how effective GI will be when they reach production scale.
GI is extremely specialized in the area of spatial reasoning within the context of gaming. Therefore, GI is unlikely to be able to provide the same versatility as other general world models.
There is also a significant talent competition risk associated with the employment of spatial AI researchers. There is currently a global "war" for talent in the field of spatial AI, with many companies competing for a small pool of highly skilled employees.
One of the reasons why GI avoids copyright issues related to using existing game footage to train their AI is because they are limiting the amount of additional game footage they can collect for future AI improvements.
To date, the vast majority of NPC behaviors have been based on simple decision making processes that do not include human-like spatial reasoning. In order for NPCs to be believable in today's gaming environment, they must possess the ability to understand and interact with their environment in a way that is similar to how humans interact with theirs.

Best For

AAA game studios — In order for the gaming experience to remain engaging for the long term, it is necessary to maintain a certain level of balance between the strength of the AI opponents and the skills of the human players. If the AI opponents become too difficult to defeat, then the game will eventually lose its appeal to the human players. Conversely, if the AI opponents are too easy to defeat, then the game will lack the challenge that keeps players interested.
Competitive multiplayer developers — Foundation models are pre-trained models that have already learned the patterns and relationships inherent in large datasets. These models can then be fine-tuned for a specific task by using less data than would be needed to train the model from scratch. A key advantage of foundation models is that they can be trained on large quantities of spatial-temporal data and therefore have the potential to learn how to understand and interpret the structure of a 3D environment in a way that is similar to how a human understands and interprets their environment.
AI research teams in robotics/gaming — An ideal solution to the problem of finding perfectly balanced AI opponents for online multiplayer games would be to use an AI agent that can adaptively adjust its difficulty to match that of each individual human player. This would ensure that every game played would be a fair and challenging contest regardless of the skill level of the player.
Esports platform operators — If an organization wishes to gain early access to the spatial reasoning breakthroughs that GI is working to achieve, then they should consider partnering with the company now in order to take advantage of what is expected to be a long lead-time before the technology becomes widely available.
Frontier AI investors — Currently, GI's products and services are only available to enterprise customers and not to solo developers. Organizations that wish to utilize GI's technology to improve their games should consider using alternative technologies that are currently available to them such as NVIDIA's ACE or Unity's ML-Agents.

Not Suitable For

Individual indie developers — Currently, GI's primary focus is on training AI agents that will participate in games. They do not provide any tools for creating game-generated content such as images, videos and text. Therefore, organizations that require such tools should look into alternatives such as RunwayML or OpenAI Sora.
General content creators — GI's products and services are currently only available to enterprise customers and not to solo developers.
Non-gaming enterprises — Gaming-specific models. General World Models from DeepMind or Covariant are used as general world models.
Budget-constrained teams — Premium research product. Open source alternatives such as Habitat-sim are used first.

Limits Restrictions

Product Availability: Pre-launch, H1 2026 public release planned
Access Model: Enterprise/research partnerships only
Training Data Source: Medal.tv gameplay footage exclusively
Use Case Restriction: Gaming AI agents/NPCs, simulation tools
API Access: Not available until commercial launch
Geographic Availability: US-based, global enterprise access expected
Compliance: Gaming data licensing restrictions apply

Security & Compliance

Data ProvenanceAll training data from licensed Medal.tv gameplay footage with proper rights management.

Research SecurityEnterprise-grade model access controls for frontier AI research.

Infrastructure SecurityCloud-based training pipelines with standard AI research security practices.

IP ProtectionStrategic avoidance of copyright issues through gaming data focus.

Enterprise AccessCustom contracts with NDA protections for model weights and APIs.

Audit CapabilitiesTraining data filtering software processes thousands of hours daily.

Customer Support

Channels

Enterprise/research partnershipsDirect contact for commercial dealsStrategic partnership inquiries

Hours: Business hours for partnership inquiries
Response Time: Custom enterprise support for research partners
Satisfaction: N/A (pre-commercial stage)
Specialized: Dedicated partnership teams for game studios and AI researchers
Business Tier: Custom support contracts for commercial deployments

Support Limitations

•No public support channels available

•Enterprise/research customers only

•No consumer or individual developer support

Api Integrations

API Type: No public API available. Product focused on proprietary foundation model research for embodied AI agents rather than developer integrations.
Authentication: Not applicable - no public developer API or authentication methods documented.
Webhooks: No webhook support. Research-stage product without integration capabilities.
SDKs: No official SDKs. No GitHub repositories or developer tools found.
Documentation: No API documentation available. Company website focuses on research mission, not developer resources.
Sandbox: No sandbox or testing environment. Not a developer-facing product.
SLA: No service level agreements. Research lab, not production service.
Rate Limits: Not applicable.
Use Cases: Not applicable for integrations. Focused on internal model training for spatial-temporal reasoning in games and simulations.

Faq

What is General Intuition?

General Intuition is a research lab developing foundation models of environments requiring deep spatial and temporal reasoning especially for embodied AI agents in games and simulations. The researchers at General Intuition develop world models that can generate interactive 3D frame distributions based on user actions.

How do General Intuition's AI models work?

The models developed by General Intuition use both action labels from user input (and not the raw keystroke/controller input) and 2D frame data to learn a 3D world representation. Once learned the models will generate subsequent frames based on actions taken by the user, as opposed to generating a sequence of frames as seen in sequential video generation.

What's the difference from typical code generation tools?

General Intuition uses continuous spatial-temporal data instead of using discrete tokens as seen in code generation tools designed for embodied agents working within games and real-world simulations. General Intuition emphasizes world models versus autoregressive text predictions.

Is General Intuition available for public use?

There are no commercial products or public access points to this technology. General Intuition is a research organization and does not provide any APIs, SDKs or user facing tools for access. It appears that only internal research has access to the technology.

What are the main applications?

The primary focus of General Intuition is developing world models for gaming environments and embodied AI agents which require spatial memory and temporal reasoning. An example would be an interactive simulation where the agent must remember and react to 3D spatial changes.

Is there pricing or enterprise access?

There is no public pricing structure or enterprise plan for General Intuition. Since General Intuition is a research organization it does not have the capability to offer subscription tiers, trials, or commercial deployment options.

Can developers integrate with General Intuition models?

Due to being a research organization there are no integration options available. There are no APIs, SDKs, or documentation available for external use. The research conducted by General Intuition is proprietary and not developer-focused.

What stage is the technology at?

In its early stages, General Intuition was focused on developing innovative model architectures to solve the fundamental challenges associated with spatial memory in open-ended environments. As a result, General Intuition has not been deployed into production.

Expert Verdict

General Intuition is considered a cutting edge research group within the field of embodied AI agents and world models for spatial-temporal reasoning, placing it at the forefront of AI for game and simulation applications. The problem is that since it has no commercial products, API's, or public access, there are no near term practical benefits for developers or business to be gained from this research group. A strong technical approach but completely unmarketed.

Researchers in the area of world models and embodied agents
Gaming organizations that want to keep up on new developments in spatial-temporal AI
Educational institutions researching continuous action space

!
Use With Caution

Developers who want tools to generate production code
Organizations who want deployable AI solutions
Development teams who need to improve their productivity immediately

Not Recommended For

All production workflow development
Organizations who require API's or integration
Organization who have budgetary constraints and expect an ROI
All non-research use cases

Expert's Conclusion

An excellent research lab for developing innovative embodied AI technology; however, a viable product for either a commercial or development entity today it is not.

Best For

Researchers in the area of world models and embodied agentsGaming organizations that want to keep up on new developments in spatial-temporal AIEducational institutions researching continuous action space

Research Summary

Key Findings

General Intuition is a research lab developing foundational models for embodied AI agents using spatial-temporal reasoning primarily for application in gaming and other types of interactive simulations. It uses action labels + 2D frames to create bootstrapped 3D world models and generate interactive frame distributions. There was no commercial product, API's, or public access available to evaluate - strictly a research focused organization.

Data Quality

Limited - primarily podcast transcripts and basic website. No API docs, pricing, case studies, or developer resources available. Information sparse due to research-stage nature.

Risk Factors

No commercial product or revenue model found

Research stage prior to commercialization

Proprietary technology with no transparency

Untested in production environments

Last updated: February 2026

Alternatives

•
Cursor: AI powered code editor with advanced autocomplete and agenic capabilities (autonomy level 3-4). Production ready for developer use unlike research only General Intuition. Best used by individual developers and development teams that develop software today. (cursor.com)
•
Replit Agent: full agentic coding environment creates whole applications from user input. Production ready now, while general intuition is focused on research. best for developing prototypes quickly and for non-experts who require a fully functional solution. (replit.com)
•
GitHub Copilot: leading AI pair programming tool with large-scale enterprise customers (Level 2 autonomy). mature, well-integrated, and proven in production as opposed to experimental embodied AI research. best for teams of developers using this tool for professional development. (github.com/features/copilot)
•
Anthropic Claude Code: advanced code creation through use of Claude models and 200k + context; provides more robust treatment of complex reasoning problems than spatially-oriented research. best for enterprises with large code bases and/or heavy reasoning required for development. (anthropic.com)
•
Applied Intuition Copilot: generative AI for automotive simulation scenarios (closely related to spatial reasoning). simulates test cases up to 40x faster when using natural language input versus pure games research. best for simulation intensive industries such as autonomous vehicles. (appliedintuition.com)