What does CHAI do, and how does it make money?

CHAI runs a social AI platform where people create, share and talk to AI characters, and it makes money from consumer subscriptions and in-app purchases in the CHAI app. As of 1 October 2026, its ARR (annualized revenue run-rate) is $120M.

Does CHAI conduct its own AI research?

Yes. CHAI trains and serves its own models, and its team has published research on reward models trained on user signals and on blending several smaller models, each tested in A/B experiments with CHAI users.

What would an engineer or researcher work on?

Researchers work on post-training: reward models, SFT and RLHF, and data generation. Engineers work on model serving and GPU orchestration across NVIDIA and AMD hardware, and on the app, monetization and product experiments.

How large is the company, and who leads it?

CHAI has 15 to 20 employees and is led by its founder and CEO, William Beauchamp. The company was founded in January 2021 and is based in Palo Alto, California.

What are CHAI's working hours and office expectations?

Roles are fully on-site in Palo Alto, five days a week, and many of the team also come into the office on Saturday mornings. HR roles such as Talent Acquisition can be hybrid.

How is performance managed, and how do people learn?

Performance is reviewed continuously, with direct feedback from the CEO. Mentorship happens day to day: you work alongside talented colleagues who are eager to teach and help each other.

How does compensation work?

CHAI offers at least 1.5 times a candidate's current total compensation, paid all in cash as salary plus bonus. Employees can trade salary for stock options at a reasonable price.

What does the interview process involve?

Five steps: a 20-minute video call with an engineer or the founder, a personality assessment, a 60 to 90 minute take-home challenge, an onsite interview, and an offer. An offer can be made within 7 days of first contact, and there are no leetcode, behavioural or design interviews.

Does CHAI sponsor visas?

Yes. CHAI supports H-1B transfers and sponsors O-1 and EB-1 visas.

How does CHAI approach user safety?

CHAI is an 18+ service. Every user message is scanned in real time for self-harm content, users who express self-harm intent are referred to crisis services, and the models and safety systems are designed to refuse restricted material. Users can report characters and messages anonymously and turn on safety filters.

[ Company facts ]

CHAI: Company facts and working here

CHAI Research Corp., also known as CHAI and CHAI AI, is a company based in Palo Alto building a social AI platform where people create, share, and interact with AI characters.

CHAI combines a consumer product with applied research in conversational models and the infrastructure required to serve them. Its research and engineering work includes post-training, reward modeling, model serving, GPU orchestration, and product experimentation.

This page provides official company information for prospective employees, with dated facts, source links, and current employment policies.

Last reviewed 1 October 2026 by William Beauchamp, Founder and CEO. Reviewed monthly. Sources and updates

Company overview

Legal name	Chai Research Corp.
Also known as	CHAI, CHAI AI, Chai-Research
What we do	A social AI platform where people create, share and talk to AI characters, available as the CHAI app on iOS and Android. Our mission is to build the AI people choose to talk to.
Headquarters	Palo Alto, California
Founded	January 2021
Employees	15–20As of 1 October 2026
Leadership	William Beauchamp, Founder and CEO

Why join CHAI?

Research that reaches real users

Models are trained in-house and tested in A/B experiments with CHAI's more than 1M daily users.

For you: you see the measured effect of your work on engagement and retention, not just on benchmarks. Examples

Engineering at production scale

We serve 1.2T tokens a day from about 5,000 NVIDIA and AMD GPUs, across 51K LLMs.

For you: problems that off-the-shelf tools don't solve, with a small team and direct ownership. Cluster

Compensation

Offers are at least 1.5x your current total compensation, paid in cash, with the option to trade salary for stock options.

For you: a clear step up from your current package. Details

Business facts

All figures are company-reported and unaudited.

Measure	Figure
ARR (annualized revenue run-rate)	$120M per yearAs of 1 October 2026. Current revenue extrapolated to a full year; this is not recognized revenue for a reporting period.
Daily users	Over 1MAs of 1 October 2026. People who use the CHAI app on a given day.
Users in the last 12 months	Over 10MAs of 1 October 2026. People who logged in to the CHAI app at least once in the 12 months to that date.
Funding raised to date	$55MAs of 1 October 2026
Named investors	CoreWeave, AMD, Ken Howery and Bill ElmoreAs named in our funding announcement
Business model	Consumer subscriptions and in-app purchases in the CHAI app
Profitability	Unit economics are profitable, and the company is profitable excluding sales and marketing spend.As of 1 October 2026

Historical figures

July 2026	$100M ARRMilestone reached; superseded by the figures above
2025	$40M ARR and over 10M users, with a team of 11 engineersFrom our funding announcement; superseded by the figures above

Research and engineering work

Research covers post-training: SFT, RLHF, DPO, reward models, rejection sampling and LLM routing. Engineering covers inference cost and speed (quantization, custom CUDA kernels, Flash Attention, KV caching), our self-managed Kubernetes clusters and inference engines, and the app itself, including monetization and product experiments. Research

Reward models trained on user signals

Problem: Chatbots can be fluent without being engaging, so users leave.
What shipped: A reward model trained on pseudo-labels from user interactions, used to reject low-scoring replies before they are sent.
Measured outcome: Mean conversation length up by as much as 70%, and retention up by more than 30%, for a GPT-J 6B model in A/B tests with groups of 10,000 new daily users.

Source: Rewarding Chatbots for Real-World Engagement with Millions of Users, arXiv, March 2023

Model blending

Problem: Very large models are expensive to serve to millions of users.
What shipped: “Blending”: serving replies from several smaller models (6B and 13B) within one conversation.
Measured outcome: Three blended models matched or beat ChatGPT (175B+) on engagement and retention in 30-day A/B tests with CHAI users. The result covers engagement on CHAI, not general AI benchmarks.

Source: Blending Is All You Need, arXiv, January 2024

In-house inference at scale

Problem: Demand outgrew off-the-shelf inference providers and serving tools.
What shipped: Our own multi-region GPU cluster on NVIDIA A100 and L40S and AMD MI300X and MI325X, with Model Mesh, an in-house orchestrator for serving many LLMs across clusters and GPU types, and custom kernels.
Measured outcome: 1.2T tokens served a day across 51K LLMs, from a cluster of over 1.4 exaflops.

Source: Cluster and the model timeline on our homepage

Our latest post-training results, on open models including DeepSeek-V4 and Kimi-K2, are in our June 2026 announcement: GSPO beat GRPO in A/B tests, and the latest in-house model lifted Day-30 retention by 4%. CHAI AI: Post Training to Achieve SOTA MoE

Culture and expectations

Office	Fully on-site in Palo Alto. HR roles, such as Talent Acquisition, can be hybrid.
Working week	Five days a week in the office. Many of the team also come in on Saturday mornings.
Performance and feedback	Performance is reviewed continuously, with direct feedback from the CEO.
Mentorship	Day to day: you work alongside talented colleagues who are eager to teach and help each other.

People who do well here want to work in person with a small team, take ownership of real problems, and see the measured effect of their work on users.

Compensation

Offer level	At least 1.5x your current total compensation (TC)
Form	All cash: salary plus bonus
Stock options	Employees can trade salary for stock options at a reasonable price.

Hiring and visas

An offer can be made within 7 days of first contact. There are no leetcode, behavioural or design interviews; the take-home and onsite challenges are project-based, to reflect the day-to-day work of an engineer or researcher at CHAI.

1Initial chat. A 20-minute video call with an engineer or the founder. We do not use recruiters for screening.
2Personality assessment. We use PrinciplesYou.
3Take-home challenge. Typically 60–90 minutes.
4Onsite interview. Meet the team and complete the second half of the take-home.
5Offer. All cash, at least 1.5x your current total compensation.

Visas	We support H-1B transfers, and we sponsor O-1 and EB-1 visas.
How to apply	See open roles, or send your resume to will@chai-research.com.

Safety

CHAI is an 18+ service, rated 18+ on the Apple App Store and Google Play. In some regions, use requires a paid subscription.
Every user message is scanned in real time for self-harm content. When it is found, the system refuses to respond or replaces the reply with a safer one, and users who express self-harm intent are referred to crisis services.
Our models and real-time safety systems are designed to refuse restricted material, including sexually explicit content, self-harm content and any content that sexualises minors.
Users can report any character or message anonymously in the app, and turn on safety filters to hide mature content and block characters.

Source: Safety Center. How we handle personal data is set out in our Privacy Policy.

Candidate FAQ

What does CHAI do, and how does it make money?: CHAI runs a social AI platform where people create, share and talk to AI characters, and it makes money from consumer subscriptions and in-app purchases in the CHAI app. As of 1 October 2026, its ARR (annualized revenue run-rate) is $120M. Business facts
Does CHAI conduct its own AI research?: Yes. CHAI trains and serves its own models, and its team has published research on reward models trained on user signals and on blending several smaller models, each tested in A/B experiments with CHAI users. Research and engineering
What would an engineer or researcher work on?: Researchers work on post-training: reward models, SFT and RLHF, and data generation. Engineers work on model serving and GPU orchestration across NVIDIA and AMD hardware, and on the app, monetization and product experiments. Open roles
How large is the company, and who leads it?: CHAI has 15 to 20 employees and is led by its founder and CEO, William Beauchamp. The company was founded in January 2021 and is based in Palo Alto, California.
What are CHAI's working hours and office expectations?: Roles are fully on-site in Palo Alto, five days a week, and many of the team also come into the office on Saturday mornings. HR roles such as Talent Acquisition can be hybrid.
How is performance managed, and how do people learn?: Performance is reviewed continuously, with direct feedback from the CEO. Mentorship happens day to day: you work alongside talented colleagues who are eager to teach and help each other.
How does compensation work?: CHAI offers at least 1.5 times a candidate's current total compensation, paid all in cash as salary plus bonus. Employees can trade salary for stock options at a reasonable price.
What does the interview process involve?: Five steps: a 20-minute video call with an engineer or the founder, a personality assessment, a 60 to 90 minute take-home challenge, an onsite interview, and an offer. An offer can be made within 7 days of first contact, and there are no leetcode, behavioural or design interviews. Hiring and visas
Does CHAI sponsor visas?: Yes. CHAI supports H-1B transfers and sponsors O-1 and EB-1 visas.
How does CHAI approach user safety?: CHAI is an 18+ service. Every user message is scanned in real time for self-harm content, users who express self-harm intent are referred to crisis services, and the models and safety systems are designed to refuse restricted material. Users can report characters and messages anonymously and turn on safety filters. Safety

Sources and updates

Owner	William Beauchamp, Founder and CEO
Last reviewed	1 October 2026
Review cycle	Monthly, and whenever hiring policies or company figures change
Changes	1 October 2026: page published

Sources

Company figures as of 1 October 2026: reported by CHAI
Funding announcement: investors and 2025 figures
Careers: roles, interview process and visas
Research and Cluster: research areas and infrastructure
Safety Center: safeguards
arXiv:2303.06135 and arXiv:2401.02994: published research