Pay for problems solved, not time spent
Resolution-based pricing: a flat rate for every task the agent resolves, and nothing for the ones it doesn’t.
- 1
Tell us about the work
How many requests need work done in another system, and which systems.
- 2
Agree what counts as resolved
Up front, in your words: the address is updated, the claim is filed, the refund is issued.
- 3
Pay per resolution
A flat rate for each task the agent resolves, priced against what a callback or ticket costs you today. Nothing for the ones it doesn’t.
Priced by usage, a more efficient agent earns us less. Priced by the resolution, a better agent earns us more and saves you more. It puts us on your side of the table.
Or pay as you go, from about 1¢ a task
For the tokens your agents use, from a prepaid balance, at the price of the AI model each one runs on. Tasks and chat are priced the same way. You choose the model for each agent, and each procedure can have its own: its evals show the cheapest one that still passes. Procedures can also run as steps without AI, so the agent’s model writes a reply only when a conversation leaves the script.
| Model | Input | Cached input | Output | Typical task |
|---|---|---|---|---|
Claude Haiku 5.5Anthropic | $0.15 | $0.015 | $0.75 | 1¢ |
| Anthropic’s fastest and lowest-priced model, for simple, well-defined tasks. Cached input $0.015. | ||||
GPT-6 LunaOpenAI | $0.15 | $0.015 | $0.75 | 1¢ |
| OpenAI’s lightest GPT-6, for simple, high-volume tasks. Cached input $0.015. | ||||
DeepSeek V4.1 FlashDeepSeek | $0.45 | $0.009 | $1.80 | 1¢ |
| DeepSeek’s fast, low-priced model, and it sees screens. Runs on DeepSeek’s servers in China. Cached input $0.009. | ||||
DeepSeek V4 ProDeepSeek | $1.98 | $0.066 | $5.94 | 6¢ |
| DeepSeek’s strongest thinker, for chats at a low price. Text only: no browser or desktop tasks. Runs on DeepSeek’s servers in China. Cached input $0.066. | ||||
GPT-6.1 SolOpenAI | $3 | $0.15 | $15 | 10¢ |
| Close to Astra on computer use, for a fraction of the price. Cached input $0.15. | ||||
Claude Sonnet 5.5Anthropic | $3 | $0.15 | $15 | 15¢ |
| Fast and accurate on screens. The best value for most support tasks. Cached input $0.15. | ||||
Kimi K3.1 ProMoonshot AI | $3 | $0.30 | $15 | 16¢ |
| Moonshot’s newest Kimi: thinks hard and sees screens. Runs on Moonshot’s servers in Singapore. Cached input $0.30. | ||||
Kimi K3.1 Pro HighspeedMoonshot AI | $6 | $0.30 | $30 | 27¢ |
| Kimi K3.1 Pro at two to three times the speed. Runs on Moonshot’s servers in Singapore. Cached input $0.30. | ||||
Claude Opus 5.5Anthropic | $6 | $0.30 | $30 | 31¢ |
| Anthropic’s strongest at computer use, for long flows and fiddly settings. Cached input $0.30. | ||||
Gemini 3.8 FlashGoogle | $2.25 | $0.225 | $11.25 | 32¢ |
| Google’s quick model, for simple, well-defined tasks. Cached input $0.225. | ||||
GPT-6 AstraDefaultOpenAIDefault | $15 | $1.50 | $75 | 58¢ |
| OpenAI’s most capable model. What Eager has always run. Cached input $1.50. | ||||
Claude Fable 5.1Anthropic | $15 | $0.375 | $75 | 71¢ |
| Anthropic’s most capable model, for the hardest multi-app problems. Cached input $0.375. | ||||
Gemini 4 ArgonGoogle | $6 | $0.30 | $30 | 86¢ |
| Google’s most capable model, for long, multi-step problems. Cached input $0.30. | ||||
US dollars per million tokens. A typical task is 10 steps in a browser; yours depend on their length and screens. Claude models also charge for input written to their prompt cache, at 1.25× the input price ($0.188, $3.75, $7.50, $18.75), as Anthropic does. On Claude Haiku 5.5, a call whose input is over 100K tokens costs $0.75 input and $3.75 output. On GPT-6 Luna, a call whose input is over 272K tokens costs $0.30 input and $1.125 output.
What the levels mean
How much the model thinks before each step. Thinking more uses more tokens.
- LowDefault
- Fast and direct. Right for most support tasks.
- Clear, well-trodden tasks: printers, Wi-Fi, installs, account settings.
- Medium
- Thinks more before each step, for fiddly settings and longer flows.
- Long forms, settings buried a few screens deep, setups that take many steps.
- High
- The most careful, for tricky screens and multi-app problems.
- Unfamiliar or cluttered apps, and problems that cross several apps. The slowest of the three.
$20 free credit when you sign up, for tasks and chat on any model, with no card. Every model includes the console, the API, guardrails and invite links with your team’s name.
Prepaid, with auto top-up
What your agents use comes out of a balance you add to by card. Nothing is charged that you didn’t set up.
- Add funds by card
- From $10 at a time, through Stripe. Your balance doesn’t expire.
- Auto top-up, if you want it
- When your balance drops below $10, we add $50 from your saved card. You choose both amounts, and it stays off until an owner turns it on.
- A monthly limit
- Auto top-up never charges more than your limit in a calendar month (UTC). You set it; it’s $500 unless you change it.
How usage is counted
You pay for the agent’s steps. When it stops to ask you something, nothing is used.
- What counts as a resolution?
- We agree on it with you before you start, for the tasks you send the agent: the change the customer or rep asked for is made, without your team stepping in. Tasks it doesn’t resolve cost nothing.
- What are tokens?
- What the model reads (the task, the screen at each step, the conversation so far) and what it writes back, counted the way its provider counts them. Input it has already read in a task or chat comes from its prompt cache, at the cached price. Each task and each chat reply is rounded up to the next cent, once.
- Is chat priced the same way?
- Yes. A chat reply is paid by its tokens at its agent’s model price. A browser or computer task the reply starts is paid by its own tokens, like any task.
- What does talking to Coach out loud cost?
- 7.5¢ a minute, by the second, for the voice with OpenAI’s GPT-Live, or 12¢ a minute with an ElevenLabs voice (each agent picks one). What it does for you is paid by its tokens, like any chat reply.
- What do phone calls cost?
- 7.5¢ a minute, by the second, with OpenAI’s GPT-Live answering, or 12¢ a minute for an agent with an ElevenLabs voice. What the agent does on a call is paid by its tokens, like any chat reply. You bring your own Twilio numbers, and Twilio bills you for them and their minutes.
- What is AI Review?
- After each run that followed one of an agent’s procedures, a small model checks the run against the steps, so failures and better ways reach your team as escalations. It’s paid by its tokens at that model’s price, usually a cent or two a run, and you can turn it off for any agent.
- What isn’t counted?
- Time waiting for your answer or approval: the model isn’t running, so it uses no tokens. If a restart or update on our side ends a task, what it was charged goes back to your balance.
- Which model should I pick?
- Start with the default, GPT-6 Astra, or Claude Sonnet 5.5: either handles most tasks. If an agent struggles with a screen, move it to a more capable model or a higher level. A higher level thinks more before each step, so it uses more tokens. You set both for each agent in the console, under Agents, and can change them any time.
- What happens if the balance runs out mid-task?
- A task is paid as it runs. With auto top-up on, it adds funds and the task carries on. Without it, or once it has reached its monthly limit, the task stops after a short grace period, which isn’t charged, and says why. New tasks can’t start until you add funds.
- How does the monthly limit work?
- It’s the most auto top-up will charge in a calendar month (UTC). If the next top-up would go over it, we charge only what’s left under the limit, or wait until next month. Funds you add yourself don’t count toward it.
- How many tasks can run at once?
- 1 at a time until you first add funds, then up to 10 across your team. Any number of computers can be connected. If you need more, write to us.
- Can I get a refund?
- Yes. Your balance doesn’t expire, and we refund the funds you added that are left, on request, to the card they came from. Free credit and other credit we gave you aren’t refundable. Write to support@guidinghand.ai.
- Can I try it first?
- Yes. Sign up free and you get $20 of credit for real tasks and chat, on any model, with no card.
Already have an account?
Sign in to see your balance, add funds and set up auto top-up.