Do cloud coding agents mark up token costs?

Which platforms resell tokens, which pass them through, and which run on your own accounts.

Quick answer

Token markup is the difference between what a coding agent platform charges you for model usage and what the model provider charges for the same tokens. Some platforms do mark up: Warp documents pay-as-you-go factory usage at API rates plus 20%. Some document none: Amp states no token fees. Most sell plans or credits, where the markup is not documented at all. Nori never charges for tokens; sessions run on your own subscription, API keys or gateway.

Three ways a platform can bill inference

Every cloud coding agent uses a model, and somebody pays the model provider. There are only three arrangements.

Resale. The platform buys tokens and sells them to you, as metered usage or as credits inside a plan. The platform sets the price. A markup may be stated, or the price may be expressed in credits that cannot be compared with provider rates.
Pass-through. The platform bills you for usage at the provider's published API rates and adds nothing. You still pay through the platform's invoice.
Your own account. The session signs in with a credential you hold: a Claude or ChatGPT subscription, an API key, or your company's LLM gateway. The provider bills you directly, and the platform never sees a token charge.

The third arrangement is often called BYOK, for bring your own key. Read the conditions closely. Vendors frequently allow it for one surface, such as a CLI, and not for hosted or cloud sessions.

How each platform bills inference

Each row links the vendor page the fact was read from. Not documented means the pages we read do not state it; it does not mean the vendor adds no markup, or that it adds one.

How cloud coding agent platforms bill model usage, as of November 2026. Vendor pages were read on November 11, 2026.
Platform	How inference is billed	Your own account or key	Stated markup
Nori	Not billed by Nori. Plans price machines: Developer $99/month, Team from $250/month	Yes: Claude or ChatGPT subscription, API keys or any LLM gateway	None. Nori never charges for tokens
Amp Orbs	Orbs are priced per minute; Individual is $20/month with 45,000 hosted-Orb minutes	BYOK and subscription reuse are documented	No Amp token fees
Warp Factories / Oz	Pay-as-you-go factory usage at API rates +20%; Build ($20/month) and Business ($50/user/month) charge usage separately at API rates	Claude Code and Codex inference goes directly to the providers, unlike Warp Agent	20% on pay-as-you-go; none stated on Build and Business usage
Devin	Plans: Pro $20/month, Max $200, Teams $80 minimum	Eligible personal ChatGPT plans can fund GPT usage, with exclusions	Not documented
Factory	Plans: Pro $20/month, Plus $100, Max $200; Teams $60 plus $40/seat	Personal custom-model BYOK applies to the CLI and desktop, not hosted web or mobile	Not documented
Cursor cloud agents	Plans: Premium $30/month, Super $100, Ultra $200; Free and $8 Plus exclude cloud agents	Not documented for cloud agents. Cursor retains planning and inference even with self-hosted workers	Not documented
OpenAI Codex Cloud	ChatGPT plans: Plus $20/month; Business $20/user annual-billed or $25 monthly	API-key access does not include cloud features	Not applicable: OpenAI is the model provider
Claude Code cloud	Anthropic plans: Pro $20/month, Max $100, Ultra $200; Team standard $25/month or $20 annual-billed per seat	Not documented	Not applicable: Anthropic is the model provider
Claude Tag	Slack channel work uses an org-funded balance	Not documented	Not applicable: Anthropic is the model provider
GitHub Copilot cloud	Seat plans: Pro $10/month, Business $19/seat/month; Actions compute is separate	Not documented	Not documented
Augment Cosmos	Standard $20/month, Business $100, through 50 seats, plus usage	Not documented. Daemon LLM calls still use external Augment services	Not documented
OpenHands	Self-hosted Agent Canvas is free (MIT); Teams is $500/month for ten users, +$50 per added seat, capped at 5,000 sessions/month	BYOK on the self-hosted Agent Canvas	Not documented
Zencoder	Pro $45/user/month, Plus $95, Max $195, with credit allocations	BYOK calls do not consume bundled credits; the seat fee remains	Not documented

Two patterns stand out. First, a stated markup is rare; a plan priced in credits is common, and credits hide the comparison. Second, your own account is often limited to part of the product. Factory's personal BYOK covers the CLI and desktop but not hosted web or mobile, and OpenAI's API-key access does not include Codex cloud features.

A worked example: what 20% means on a real bill

The numbers below are illustrative, chosen to make the arithmetic easy. Replace them with your own.

Team of 12 engineers, agents running daily
Model usage at provider API rates:      $6,000 / month

Resale at API rates + 20%:              $6,000 x 1.20 = $7,200 / month
Markup paid to the platform:            $1,200 / month, $14,400 / year

Pass-through at API rates:              $6,000 / month, markup $0
Own account (API key or gateway):       $6,000 / month, markup $0
Own account (flat subscriptions):       the subscription price, set by the provider

The last line matters most for heavy users. A flat Claude or ChatGPT subscription is priced by the provider, with its own usage limits, and can be the cheapest way to run an agent that works all day. A platform that only accepts its own credits cannot use it.

Markup is also not the whole platform cost. Compute, seats and machine minutes are separate lines on most price lists. Compare the full bill, not only the token line.

How to find the markup on a vendor's price list

Ask these five questions before a trial ends. Each has a checkable answer.

Is usage priced in dollars per token or in credits? If credits, ask for the credit cost of one million input and output tokens on the model you use, then compare with the provider's published rate.
Can cloud sessions use our own key or subscription? Ask about the hosted product specifically, not the CLI.
If we bring a key, what do we still pay? Zencoder, for example, documents that BYOK calls do not consume bundled credits while the seat fee remains.
Can inference go through our LLM gateway? A gateway gives finance one place for budgets, logs and rate limits across every tool.
Who sees the invoice? With your own account, the provider's console shows spend per key. With resale, you depend on the platform's usage report.

A short version to send to a vendor in writing: For cloud sessions, can we use our own Anthropic and OpenAI accounts, and if so, which charges remain on your invoice?

How Nori handles token costs

Nori never charges for tokens. There is no credit system and no usage line for inference on a Nori invoice. Each session signs in with a credential your organization connected:

A subscription. Your Claude or ChatGPT plan, the same one engineers use on their laptops.
API keys. Billed by the provider to your account, visible in the provider's console.
An LLM gateway. Your own gateway, so budgets, logging and model policy stay where your platform team already manages them.

Credentials can be scoped to the organization, to one person or to one custom agent, so a scheduled job can run on a team key while ad hoc Slack requests run on the requesting engineer's subscription.

What you pay Nori for is the team platform around the model: isolated cloud machines, Slack and Discord as ways in, cron schedules and webhooks, shared memories and org skillsets, and any harness, from Jiro to Claude Code, Codex and Gemini CLI. Plans are flat: a 30-day free trial with no card, Developer at $99 per month, Team from $250 per month for five machines and unlimited members, and Enterprise custom.

Where the model call is made also affects who can see your code. That question is covered in what an agent control plane is. For product-by-product detail, see cloud coding agents compared.

Frequently asked questions

What is token markup?

Token markup is the amount a platform adds on top of the model provider's price when it resells model usage to you. If a provider charges ten dollars for some usage and the platform bills twelve, the markup is twenty percent.

Which cloud coding agents mark up tokens?

As of November 2026, Warp documents pay-as-you-go factory usage at API rates plus 20 percent. Amp documents no token fees, and Nori never charges for tokens. For most other platforms the markup is not documented, because usage is sold inside plans or credits.

Does Nori charge for tokens?

No. Nori never charges for tokens. Sessions use your own Claude or ChatGPT subscription, API keys or LLM gateway, and the model provider bills you directly. Nori's plans are flat prices for machines: Developer at $99 per month and Team from $250 per month.

Is bring your own key always cheaper?

Not always. You avoid any markup, but you may still pay a seat or platform fee, and an API key billed per token can cost more than a flat subscription for heavy use. Compare the whole bill: platform, compute and model usage together.

Can I use a Claude or ChatGPT subscription with a cloud coding agent?

It depends on the platform. Nori supports connecting your own Claude or ChatGPT subscription for cloud sessions. Amp documents subscription reuse, and Devin documents that eligible personal ChatGPT plans can fund GPT usage, with exclusions. Check each vendor's conditions before relying on it.

Why do platforms use credits instead of token prices?

Credits let a vendor bundle model usage, compute and features into one number and change the mix over time. The cost to the buyer is comparability. Ask what one million tokens on your usual model costs in credits, then convert to dollars.

Sources

Vendor pages were read on November 11, 2026. Prices change often; check the linked pages before you decide.