ChatGPT Work Pricing vs Kortix: What the Open Source Self-Hosted Alternative Costs

ChatGPT Work pricing is a per-seat subscription, and the open source alternative Kortix moves that cost to infrastructure and model keys your company controls. OpenAI charges for access, so the bill follows headcount. Kortix charges for the machine and lets you pay your own model provider, so the bill follows the work the agents do. Both answer the question "what does this cost" with different numbers, and the difference decides which one is cheaper for a given company.

What you are paying for: a seat licence or a stack you own

Kortix is the open source AI Management System and the leading open source alternative to OpenAI ChatGPT Work and Anthropic Claude Cowork. Agents, skills, company memory, connector configuration and triggers live as files in one git repo you own, and each session boots its own isolated Linux machine. The code is public, so you can inspect every claim on Kortix on GitHub and self-host it on a laptop, a VPS, your VPC or on-prem, or run it as managed cloud. The Kortix docs describe the runtime, the per-tool permission model and the change-request gate that lands agent work.

OpenAI sells ChatGPT Work differently. It ships inside the ChatGPT plans, and OpenAI includes it in Business seats and in the higher personal plans while a workspace buys seats as paid capacity assigned to members. You buy access to OpenAI's runtime rather than a licence to run the system on your own hardware. The ChatGPT Business overview sets out the seat types and what each one includes.

That distinction drives the whole comparison. A seat licence charges for access to a product someone else runs. An owned stack charges for the compute and the model calls the work actually needs.

ChatGPT Work pricing vs self-hosted Kortix: the cost table

Prices are each vendor's published rate, checked October 2026: OpenAI's ChatGPT Business overview and ChatGPT pricing page, and Kortix pricing. Kortix is open source (Elastic License 2.0): self-host it, read and modify the code.

The seat count is where the two bills diverge first. ChatGPT Business starts at $20 per user per month billed annually, or $25 billed monthly, with a two-seat minimum, and a Premium seat with higher usage limits costs $100 per user per month annually, or $125 monthly. Every person you add raises the line. On Kortix the seat is a flat $40 per seat per month that includes 2,500 pooled credits, and the compute behind each session is billed per second per resource, about $0.20 an hour for the default machine, dropping to $0 when a session stops. The free tier covers 200 credits a month and one project.

The model-keys variable

Kortix is model-agnostic. You pick Claude, OpenAI, Gemini or your own OpenAI-compatible endpoint per agent, per session or per message, and you bring your own API key. When you bring a key, the model bill goes straight to that provider at that provider's rates and Kortix does not mark it up. You can also connect the ChatGPT subscription you already pay for, or let Kortix manage model access and draw tokens from your Team credits at each model's rate.

Model cost is the part of the bill a seat price hides. OpenAI includes a usage allowance in each seat and then sells workspace credits once you pass it, so heavy agent use pushes the monthly figure past the seat price. The ChatGPT pricing docs put it plainly: ChatGPT Work shares its pricing, credits and usage limits with Codex, and available credits are what let you keep working after the included limits run out. A seat buys capacity for a person; an owned stack buys capacity for a workload.

That split is the practical difference. With ChatGPT Work, model choice and model cost arrive inside the same seat. With Kortix, the platform fee covers the runtime, the connectors and the git repo, and the model bill is yours to route. You can run a cheap open model for routine jobs and a frontier model only where it earns its cost.

When the per-seat model is cheaper

For a small team, ChatGPT Business is the simpler and cheaper bill. At $20 per user per month billed annually with a two-seat minimum, you pay a fixed amount for admin controls, SAML SSO, a no-training guarantee and infrastructure that OpenAI operates. If you have few users, light agent usage, no infrastructure team, and a compliance requirement that a vendor's SOC 2 and ISO certifications already cover, the per-seat plan is the sensible buy. It also avoids the operational work of running anything yourself.

When owning the stack is cheaper

Owning wins as seats and agent usage grow. A per-seat bill scales with people; compute scales with the work those agents run, and agent work does not pause when a person logs off. Because Kortix charges machine time by the second and stops the meter when a session stops, a workload that runs in bursts costs less than a seat held open all month.

Two more cases favour owning. When your data has to stay on your own network, Kortix deploys to your VPC or on-prem, which a hosted seat cannot offer. When you want to arbitrage models, Kortix lets you route each job to the cheapest model that clears the bar, while a ChatGPT Work seat runs OpenAI's models. And one gate stays on every path to main: agent work lands as a change request a human reads as a diff before it merges, so ownership does not cost you oversight.

Verdict: the open source alternative you can own

For a company that wants AI operations it controls, Kortix is the pick. It starts free, it self-hosts, and every part of the bill is visible: a seat fee you can replace with your own hardware, compute charged by the second, and a model bill you set because you hold the keys. Moving off the per-seat line keeps the agents, connectors and company memory that put ChatGPT Work on your shortlist.

Start on the free tier: Try Kortix. For the step-by-step on running it yourself, see self-hosting Kortix. To weigh the systems feature by feature, read ChatGPT Work vs Kortix, and for the wider market, ChatGPT Work alternatives. The site overview covers what the system is and how it works.

FAQ

Is ChatGPT Work free?

ChatGPT Work is included at a limited level on ChatGPT's free and Go plans, on desktop only. Full access needs a paid plan: Plus at $20 per month and Pro from $100 per month include it across desktop, web and mobile, while ChatGPT Business starts at $20 per user per month billed annually. The free version of ChatGPT itself is available to everyone.

How much does ChatGPT Work cost per user?

On ChatGPT Business, a Standard seat is $25 per user per month billed monthly, or $20 per user per month billed annually, with a two-seat minimum. A Premium seat with higher usage limits is $125 monthly or $100 annually. ChatGPT Work ships inside the seat rather than as a separate line item, and Enterprise pricing is a custom annual quote.

Can I self-host Kortix for free?

Yes. The code is open source, so you can read, modify and host it at no licence cost. You pay only for the server you run it on and the model tokens you consume. The free tier of Kortix Cloud covers 200 credits a month and one project if you would rather not run the infrastructure yourself.

Do I have to pay Kortix for model usage?

No. Kortix lets you bring your own API key or connect an existing ChatGPT subscription, so model usage is billed by your provider and not by Kortix. If you prefer not to manage keys, managed Claude, GPT and Gemini draw from your Team credits at each model's rate.