Chai is an AI chat application for entertainment and role-playing. Users can have interactive conversations with various preset or user-created AI characters, emphasizing rich character settings and immersion.

Chai

Chai’s core parameters and statistics

Chai is a role-playing AI platform for entertainment and social interaction. It is officially positioned as a "Social AI Platform" - users can create, share and talk to AI characters on the platform. Unlike Replika, which emphasizes emotional companionship, Chai focuses more on entertainment and role-playing immersion. The platform only provides mobile applications for iOS and Android, and the web version is in "Coming Soon" status.

Projects	Public Information
Official positioning	Social AI Platform — AI Characters & Stories
Product form	Mobile role-playing AI chat application
Founder	William Beauchamp (former quantitative trader, Cambridge University background)
Company Headquarters	Palo Alto, California, USA
Established	2021
Cumulative financing	More than $55M (CoreWeave, AMD, etc. investments)
Latest valuation	~$2.4 billion (Q1 2026, third-party estimate)
Annualized Revenue	~$70–80M (2026)
Daily active users	~1 million (public data in 2025)
Total users	Over 10 million
GPU Cluster	1.4 ExaFLOPs, 5,000 GPUs (A100, L40S, AMD Mi325x/Mi300x)
Daily processing Token	1.2 trillion
Number of inference models	51,000+ unique LLMs
Latest model architecture	235B A22B Mixture-of-Experts
Supported platforms	iOS, Android (Web coming soon)

Positioning Difference: Chai does not make a general AI assistant or a psychological companionship tool. Instead, it positions itself as an entertainment social platform with "user-generated AI characters". In an interview with Latent Space, founder William Beauchamp compared it to the "AI version of TikTok" - users constantly swipe to discover new characters and interact with them, thus forming a cycle of consumption and creation of UGC content.

Infrastructure scale: The GPU cluster of 1.4 ExaFLOPs is at a high level among AI consumer applications. Processing 1.2 trillion Tokens every day shows that the density of conversation traffic is extremely high. The platform serves more than 51,000 unique LLMs at the same time, which is much higher than most conversational applications. It is the combined result of third-party submitted models and internal experimental versions in the Chaiverse ecosystem.

Chai’s users and market recognition

Chai's market recognition is reflected in three aspects: continuous capital investment, rapid growth in user scale, and the gradual formation of the developer ecosystem.

Financing pace and capital background: A $2M seed round was completed in June 2023. In September of the same year, GPU cloud vendor CoreWeave made a strategic investment with a valuation cap of $450 million. A new round of financing was completed in January 2024, and CoreWeave and AMD made additional investments in 2025, bringing the total financing to more than $55M. Q1 2026 third-party estimated valuation is approximately $2.4 billion. The investors include GPU computing power providers and chip manufacturers, indicating that infrastructure capabilities are the core support of the financing story.

User Scale and Growth: Wikipedia recorded approximately 1 million daily active users in 2025, with over 10 million total registered users. Data from the official website for 2026 shows annualized income of approximately $70–80M. Chai has moved from an early stage of experimentation to a stage of scaled growth.

Chaiverse Developer Ecosystem: Chaiverse, launched in 2023, is Chai’s differentiated capability from most role-playing AI applications—third-party researchers and developers can submit their own LLMs and evaluate model performance through real user interaction data, and the winners receive cash rewards. This "crowdsourced model optimization" mechanism allows Chai to continuously iterate on conversation quality without relying on a single underlying model.

Controversy and Challenges: The suicide of a Belgian user in 2023 triggered extensive discussions on the safety of AI chat; the country-based tiered charging strategy in January 2026 triggered a user backlash; the unnotified Token limit in February of the same year caused the conversation of some paying users to be interrupted for several hours to a week. These incidents show that there is no simple solution to the challenges of security governance and user experience in AI role-playing products.

Chai’s cost advantage

Chai's cost structure is centered on C-end consumer subscriptions, which is completely different from the cost logic of API-based AI products or enterprise-level AI platforms.

C client/individual users:

• Free version: Ad-supported, there is a limit on the number of messages per day (the specific value is subject to real-time display in the app), access to the basic character library, and some advanced characters require payment to unlock.
• Premium (approximately $14.99/month or $99.99/year): Unlimited messages, no ads, priority access to new roles, faster response times, advanced customization options. The annual payment discount is about $8.33/month, which is a saving of about 44% compared to the monthly payment.
• Ultra Tier: Some reports mention higher-priced subscription tiers that offer more perks. The specific pricing is subject to the real-time page of the app store.
• Historical event: In January 2026, Chai launched a country-based tiered payment strategy due to computing power cost pressure. CEO William Beauchamp publicly stated that "the increase in computing power bills is the core reason driving this change."

Developer/API: Chai does not provide API services to external parties. Unlike competing products like Character.AI, Chai does not have an API pricing tier for developers. Although the Chaiverse platform allows third parties to submit models, it does not commercialize them through API calls.

Enterprise/Private: Undisclosed. Chai currently focuses on the C-end consumer market and has not disclosed enterprise-level deployment or privatization plans.

Cost comparison with competing products:

Dimensions	Chai	Character.AI	Replika
Free version message limit	Yes (daily limit)	Yes (daily limit)	Yes (daily limit)
Premium monthly	~$14.99/month	~$9.99/month	~$7.99/month ($49.99 annually)
Premium annual fee	~$99.99/year	~$99.99/year	~$49.99/year
API service	Not provided	Provided (c.ai API)	Not provided
Ad-supported	Free version with ads	Free version without ads	Free version with ads
Core positioning	Role-playing entertainment	Universal character dialogue	Emotional companionship

Chai’s subscription price is mid-to-high in the same category. Its cost support comes from two aspects: the infrastructure overhead brought by the 1.4 ExaFLOPs cluster, and the incentive cost of third-party model access in the Chaiverse ecosystem. For users who only experience basic functions, the limit of the free version needs to be carefully evaluated to see whether it meets daily use needs.

Main functions of Chai

Chai's functional system is designed around four sections: "Character Discovery - Character Interaction - Character Creation - Ecological Feedback".

• Character discovery and recommendation flow: The application displays AI characters in a vertical sliding card flow similar to TikTok. Users can quickly browse different characters by sliding up and down, and click to enter the conversation. The recommendation algorithm is based on role popularity, user preferences and interaction history. This design significantly reduces the decision-making cost of "finding a character from scratch" and turns browsing itself into a part of the entertainment experience.
• Character Creation Tool: Code-free tools allow users to create custom characters - setting names, avatars, opening lines, personality traits, backstory and dialogue styles. After creation, the role immediately enters the recommendation flow for other users to discover. Advanced users can write more detailed personality settings, down to speaking habits, knowledge boundaries and emotional tendencies.
• Immersive character dialogue: The dialogue engine is based on a self-developed LLM cluster and adopts "Model Blending" technology - dynamically routing the reasoning results of multiple models in a single dialogue to maintain character consistency, style stability and emotional rendering. Good character setting and model mixing can allow users to build a sense of immersion within 3–5 rounds of dialogue.
• Character Voice (TTS): The character voice function introduced in v4.8 gives the character voice characteristics and supports a variety of voices and emotional tones. Voice output significantly enhances the sense of presence in emotional dialogue scenes, and is currently only available for some roles and Premium users.
• Chaiverse Developer Ecosystem: Independent of user-side functions, Chaiverse allows third-party developers to submit LLM for A/B testing and evaluation through real user interaction data. The winning model receives a cash prize and is integrated into conversation routing. This mechanism extends model optimization from internal teams to external communities, forming a continuous model supply pipeline.
• Community interaction and evaluation system: Users can rate, comment and share characters. Popular characters get higher exposure in recommended streams

. Character creators accumulate community influence through character popularity, forming a creative incentive cycle.

Expert View: There is an implicit synergy in Chai's functional design - every new character generated by the character creation tool is simultaneously an increment of content for the recommendation stream, test material for the dialogue engine, and real scene data for Chaiverse model evaluation. In other words, the behavior of users creating roles simultaneously generates the triple value of content supply, model testing and ecological incentives. This is a closed structure that cannot be replicated by pure chat tools or single-role products.

Chai’s model and version evolution

Model route evolution

Phase	Time	Model Plan	Description
Initial stage	2021–2022	GPT-J	Quickly verify character dialogue scenarios based on open source models
Self-research transition	2023	Self-developed model based on user feedback	Start training proprietary model with RLHF + DPO
Model Blending	2024	Model Blending (multi-model integration)	Published the paper "Blending Is All You Need", integrating multiple LLMs at the conversation level
MoE large-scale deployment	2025	235B A22B Mixture-of-Experts	Deploy large-scale MoE architecture to improve both performance and efficiency
Continuous optimization	2026	51K+ model cluster + intelligent routing	Ultra-large model cluster + intelligent routing, average daily processing 1.2T Token

Application version evolution

Version	Time	Major Changes
v4.8	~2026-06	New character voice (TTS) function; custom character tool enhancements; dialogue model upgrade
v4.5	~2025-12	Character creation tool optimization; dialogue model underlying upgrade; UI interaction improvement
v4.0	~2025-06	New recommendation flow UI revision; character search and classification system redone; dialogue stability optimization
v3.x series	2024	Introduction of Premium subscription tier; Chaiverse model evaluation mechanism online; basic character voice experiment

The above version date is based on the official app store update record. There is currently no official accurate release log.

Chai’s technical advantages

Chai's technical architecture revolves around the three main lines of "ultra-large-scale inference cluster + model hybrid routing + continuous optimization driven by user feedback".

Hyperscale Inference Cluster: Chai operates a 1.4 ExaFLOPs GPU cluster containing 5,000 GPUs from NVIDIA (A100, L40S) and AMD (Mi325x, Mi300x). 1.2 trillion Tokens are processed every day, serving more than 51,000 unique LLMs. The cluster can run multiple versions of multiple models simultaneously for A/B testing, which is not possible under a single model architecture. Chai's self-developed inference engine has been deeply optimized based on vLLM - including custom CUDA kernels and computationally efficient attention approximation. According to officials, the performance improvement is close to an order of magnitude.

Model Blending: Chai published a paper "Blending Is All You Need" in 2024, proposing a technical idea to replace a single very large model - by integrating the inference results of multiple smaller models and dynamically mixing them at the conversation level, achieving conversation quality close to or exceeding that of large models. The engineering advantages are: individual models in the hybrid pool can be updated independently without affecting the overall service; the most appropriate model combination can be routed according to the conversation topic; the training and deployment costs are much lower than training a single large model with the same effect.

Continuous optimization driven by user feedback: Chai’s technical iteration relies on two data entities. One is the RLHF/DPO training pipeline - implicit feedback in user conversations (duration, response rate, whether to continue the conversation) is used to continuously fine-tune the model. The second is the Chaiverse ecosystem - models submitted by third parties undergo A/B testing with real users and the winners are integrated into the production-bound model mixing pool. This "crowdsourcing optimization" mechanism makes model progress independent of the size of the internal team.

Summary of causal chain:

• Cluster scale → Support multi-model parallel deployment → Implement model hybrid routing → Improve dialogue quality and role consistency
• Chaiverse Ecology → Continuous input of external models → Continuous optimization of the model mixing pool → Improved user participation
• User feedback data → RLHF/DPO training pipeline → Model iteration → more accurate character response

How to use Chai

Chai's usage entrance is concentrated on the mobile terminal, and the operation path is direct.

How to use	Suitable for everyone	Key steps
iOS App	iPhone/iPad users	Search "Chai: Social AI Platform" in App Store or download through the official website link; after installation, register an account to start
Android App	Android users	Search "Chai" on Google Play or download through the official website link; the installation process is the same as iOS

Typical usage process:

1. Character discovery: After opening the app, enter the character recommendation flow (swipe the card vertically), and swipe up and down to browse different characters. Filter by category (anime, history, originals, games, etc.) and popular tags.
2. Start conversation: Click on the character card you are interested in to enter the conversation interface, and send a message to get the character's reply. Free users need to be aware of daily message limits.
3. Create Character: Click the Create button, fill in the character name, select an avatar (supports custom upload), write an opening statement, set character tags and detailed background stories. After creation, the role automatically enters the recommendation flow.
4. Manage Subscriptions: Check the current message balance, upgrade Premium, and manage subscription plans on the settings page.

Web version: The official website shows "Web Coming Soon", and there is currently no available entrance to the Web side.

Chaiverse (Developer): Third-party developers can submit models through the Chaiverse portal on the official website. After the models are tested and evaluated by users, they have the opportunity to be integrated into the platform and receive rewards.

Chai’s Product Pricing

Chai uses an ad-supported free version + subscription-based freemium model, with pricing handled as in-app purchases through the App Store and Google Play.

Free version: Users can download and use it for free. There is a limit on the number of daily messages (specifically subject to real-time display in the application), and advertisements are displayed during conversations. You can access the basic character library, and some popular or advanced characters need to be subscribed to unlock.

Premium (approximately $14.99/month or $99.99/year): Unlimited messages, no ads, priority access to new and premium characters, faster dialogue response times, enhanced custom character options, character voice capabilities. The annual payment plan is discounted to approximately $8.33/month, which is a savings of approximately 44% compared to monthly payment.

Ultra / higher tier: Some reports mention the existence of higher-priced subscription tiers that provide more feature privileges. The specific price and content are subject to real-time display in the app store.

Historical pricing adjustment: In January 2026, Chai launched a tiered charging strategy by country, and users in some countries changed from free to required subscription. In February of the same year, users were dissatisfied due to the Token limit issue, and a "Basic" free tier with advertising was launched as a remedy.

Subscription Management Reminder: All subscriptions are managed through the Apple App Store or Google Play and are automatically renewed. Cancellations must be made at least 24 hours before the end of the current billing cycle.

Application scenarios of Chai

• Entertaining Role Play: Chai's most central scene. Users can experience different story lines and world views by interacting with AI characters with rich settings - from anime characters to historical figures, from original stories to virtual lovers. Unlike traditional interactive novels or games, Chai's dialogue is completely open-ended, and user behavior determines the direction of the story. The core benefit is "infinite replayability" - each conversation with a different character is a unique entertainment experience.
• Creative Writing and Character Development: Authors and content creators use Chai's character creation tools to "prototype" fictional characters - setting up personality and background and then having conversations with it to verify how natural the character's dialogue feels and discover unexpected character traits. The Canadian writer Sheila Heti, who published a series of articles in The Paris Review based on Chai's dialogue and used the dialogue in novels, illustrates the practical usability of this scene.
• Social substitution and light companionship: Users who prefer low social pressure can obtain an entertainment experience similar to social interaction through AI characters, without having to consider social etiquette or emotional burden. This scenario partially overlaps with Replika's "emotional companionship" positioning, but Chai's entertainment orientation makes its atmosphere more relaxed, and users are expected to "play" rather than "treat."
• AI model development and evaluation (Chaiverse): For AI researchers and developers, Chaiverse provides a unique environment for real user evaluation. Submitted models receive instant feedback on a platform with millions of DAUs - user interaction behavior (whether to continue the conversation, conversation length, emotional response) itself is a natural evaluation indicator of model performance.

Applicable groups of Chai

• Role-playing enthusiasts and entertainment users: Users who like immersive role-playing, interactive novels or simply want to "play" with AI characters. Chai’s character richness and dialogue quality are a definite plus in this demographic. Not suitable for: Users who seek serious literary depth or require accurate historical facts. The fictional nature of character dialogue may lead to factual deviations.
• Content Creators and Writers: Writers and creative workers who need to verify character design, explore dialogue styles, and find inspiration. Character creation tools provide quick prototype testing capabilities. Not a good fit: Creators who need a complete narrative framework or publication-grade content quality, Chai is no substitute for professional writing tools and human editors.
• AI Researchers and Developers (Chaiverse Users): Researchers who want to test and optimize LLM in real user contexts. Chaiverse provides real user interaction data and A/B testing infrastructure. Not suitable: Developers who require API commercial authorization, Chai does not provide API services for developers.
• Institutional Users Sensitive to AI Safety and Ethics: Chai's content safety policy has come under increased scrutiny following the suicides in Belgium. It is not suitable for deployment by organizations that have strict requirements for AI security compliance without a clear security audit.

Summary and Outlook

It provides competitive solutions in its field, and its core value lies in lowering the threshold for AI use in this field.

Current limitations: Some advanced features require paid subscription, and the free version has function or usage restrictions; specific technical details and performance benchmarks have not yet been fully disclosed.

Related tools: deepseek, chatgpt

How to use Chai

• Web client: You can use it by visiting the official website and registering an account. Most functions do not require installation.
• API access: Provides RESTful API, developers can obtain the API Key and integrate it into their own applications.

Version Info

• Chai v4.8 ：No official precise date yet; new character voices and custom character enhancements. （~2026-06）
• Chai v4.5 ：No official precise date yet; optimize character creation tools and dialogue models. （~2025-12）