Add-on · Beta
Tracing
The step-by-step console view of each task’s retained record. All events included during beta; included with Scale. Beta pricing, subject to change with advance notice.
$50per month
HarnessRouter Cloud pricing
- 01Every model at its listed price, no markup
- 02Pay only while your agent works. Waiting is free
- 03Bring your own keys or use ours
Estimate costDrag sliders to estimate
Without a plan$827
/ month on top-up credit at list rates
- Agent work
- $359
- Agent memory
- $25.20
- Top-up fee
- $43.12
- Model usage
- $400
Model usage on our keys$400
Included in the estimate above at listed catalog prices, with no markup.
Plans and pricingFull price included as usage.
↓ Recommended
↓ Recommended
↓ Recommended
↓ Recommended
↓ Recommended
Free
Developer
Production
Scale
Enterprise
Explore and configure
Best for your first integration
Best for apps serving real users
Best for high-volume production
Best for custom terms
$0/ month
$20/ month
$200/ month
$1,000/ month
Custom
Top up as needed
30% off top-up rates
50% off top-up rates
70% off top-up rates
Custom terms
$827 / moest. at your usage
$670 / moest. at your usage
$593 / moest. incl. top-up
$1,000 / moest. at your usage
Customvolume terms
Monthly usageresets each billing month
30 agent-work minutes
$20 of usage
$200 of usage
$1,000 of usage
Custom usage and volume
Model usageat the listed catalog price, no markup
Own keys
Included · no markup
Included · no markup
Included · no markup
Own keys or ours
Agent workper active work minute
Included, then $0.020
$0.014
$0.010
$0.006
Custom
Agent memory (Beta)per GB-hour
$0.0000840
$0.0000588
$0.0000420
$0.0000252
Custom
Concurrent tasksrunning at the same time
2
Unlimited
Unlimited
Unlimited
Unlimited
Tracing (Beta)step-by-step task record
Not included
Not included
Not included
Included (Beta)
Custom
Workspacesincluded per plan
1
1
5
Unlimited
Custom
What a task costs
A typical 3-minute text task on a mid-size model.
Task3 active work minutes · 8k input · 1k output tokens
Model$2 input · $10 output per 1M tokens · catalog price verified 2026-09-03
Example taskDeveloper plan
≈ $0.07Task total- Model usage
- $0.03
- Agent work
- $0.04
- Free wait time
- $0.00
- Tasks inside Developer’s $20
- ≈ 285
Active work is time your agent’s harness spends executing steps and tools; waiting for model responses is free. An illustration, not a measurement: these four lines are the ones on every real receipt. On your own API keys, your provider bills the Model usage line instead.
Top-up rates
Included usage is drawn first. Without a plan, or once it runs out, usage draws from your top-up credit at these rates.
- Agent work
- $0.020 per active work minuteMeasured by the second: Free wait time is $0.00.
- Agent memory (Beta)
- $0.0000840 per GB-hourAgent short-term memory, including work histories, execution trajectories, and artifacts. Metered hourly: about $0.0613 per GB-month. Beta pricing, subject to change with advance notice.
- Model usage
- The listed catalog priceNo markup on top of the upstream listing price: text, image, video, and voice models alike. See every model and price →
- Top-up credit
- $10 minimum per top-up5.5% processing fee (minimum $0.80) · valid 12 months. The fee applies to top-up purchases only — usage past a plan’s included budget is billed at that plan’s rates with no processing fee.
On the free plan, topping up lets you run models through HarnessRouter at the same listed prices, with no provider key needed.
Questions before you start.
Billing FAQExecuting a step or a tool: running code, editing files, calling an API, browsing. Waiting for a model response, waiting for you, waiting in a queue, and platform retries are Free wait time, shown on every receipt at $0.00 and split into model, user, and platform wait. Overlapping steps are measured once, by elapsed time, not once per process. Everything is measured by the second and can be recomputed from your retained task records.
Each task’s step-by-step record: what the agent did, which tools it called, what it produced, and how long each step took, next to the cost breakdown. On the free plan that record is kept for 7 days and then deleted; on every paid plan it is kept for as long as your workspace exists. Receipts and per-task totals stay in Billing regardless. Data retention is separate from agent memory: the files your tasks create count toward agent memory, the record of how they were made does not.
The console view of each task’s retained record, step by step. It is included with Scale, and available on other plans as a $50 per month add-on. All events are included during beta; pricing may change at general availability, with advance notice. The cost breakdown of every task (Model usage, Agent work, Free wait time, Task total) is free on every plan.
Every model the router can reach, from Anthropic, OpenAI, Google, DeepSeek, Alibaba Qwen, Z.ai, Mistral, Moonshot AI, and more, billed at its listed catalog price with no markup on top of the upstream listing price. Bring your own keys and your provider bills you directly instead. The catalog lists each model with its current price per million tokens. See every model and price
Yes: no card and no expiry. Explore the console, create workspaces, and configure agents at no cost. Running tasks draws from top-up credit, and your workspaces and data stay untouched whatever your balance.
Below about $20 of usage a month, topping up costs less overall, because a plan’s included usage would go partly unused; top-up credit stays valid for 12 months. From about $20 a month, a plan costs less on every rate: no processing fee (top-ups carry 5.5%, minimum $0.80), Agent work from $0.014 per active work minute instead of $0.020, longer-running tasks, and unlimited data retention instead of 7 days. Every dollar of the plan is usage.
Included usage resets at the start of each billing month and does not carry over. Billing shows how much you have left at any time. When it runs out, you choose: keep going at the same listed rates from your top-up credit, or stop until the next month. Every task shows its Model usage, Agent work, Free wait time, and Task total, so nothing on the bill is a surprise.
Your plan runs to the end of the billing month, then you drop back to the free plan; usage already incurred is still billed, and no new plan charges follow. Your workspaces and data stay with your account on the free plan, and you can export or delete them yourself at any time. Top-up credit keeps its original 12-month validity.
Self-serve plans and top-ups are card-only. Invoicing and bank transfer are available on enterprise agreements: talk to us.
- 01
Listed price means the price shown in our model catalog for the model and the upstream provider serving your request. It follows that provider’s standard published API price, and we add no markup on top of it.
- 02
We add no markup to model prices. Routing your own API keys carries no fee up to $25k a month of Model usage at listed prices; above that volume a separate routing fee of 5% of that value applies and appears as its own line on the receipt.
- 03
Longest single task on the free plan: 30 minutes, or 1 hour once you top up. Paid plans run longer tasks. current limits are shown in the console.
- 04
No separate data-transfer fee for normal app use; fair-use limits in our Terms apply.
- 05
Prices, rates, and included usage shown here are current and may change in the future. Checkout shows the final price, currency, renewal timing, and usage terms for each purchase.