AGNT Room to reason
A one-million-token window and up to 64K output tokens. Long tool transcripts, big documents and whole codebases fit in one request.
INFERENCE FOR AGENTS
A model built
for the long loop.
One million tokens of context, native tool calls, JSON output and streaming, behind an OpenAI-compatible endpoint. Priced for agents that think in dozens of steps, not one reply. Point your existing client at it and go.
Monthly plans from $5, or pay as you go with a $10 minimum top-up. $0.025 per million credits. No AGNT subscription required.
Connect over HTTPPOST
curl -X POST https://models.agnt.gg/models/v1/agents \
-H 'Content-Type: application/json' \
-d '{"name":"my-agent","allowPrepaid":true}'Returns an API key and funding link. Store the key securely. Add prepaid credit or use an eligible AGNT allowance.
1M-token context
Tool calls & JSON
OpenAI-compatible
x402 agent payments
Your agent thinks.
The bill stays small.
Repeated context costs almost nothing
Input the model has already seen bills at one credit per token. Agent loops re-send the same system prompt and history dozens of times; here that is the cheapest thing you do.
Drop-in for your client
The endpoint speaks the OpenAI chat format, with streaming, tool calling and structured output. Change a base URL and a key. Nothing else.
Keep spending bounded
Included credits come first. Extra spending requires explicit authorization and stays inside your account and key budgets. You are never charged past the cap you set on a request.
Simple prepaid pricing
Everything is priced in credits. One credit is one cached input token. A fresh input token uses 50 credits and an output token uses 200, because they cost more to serve. That is the whole price list.
| Token | Credits per token | Per million tokens |
|---|---|---|
| Cached input · context the model has seen before | 1 | $0.025 |
| Input · new context | 50 | $1.25 |
| Output | 200 | $5.00 |
$0.025 per million credits. Add $10, $25 or $50. Failed requests are not charged. Every request is settled on the tokens it actually used, never on the estimate.
Monthly plans
Buy a plan on its own — no AGNT subscription needed. Credits are already weighted to cost, so a plan is worth the same whatever you send. No rollover.
| Plan | Per month | Included credits |
|---|---|---|
| Starter | $5 | 230M |
| Pro | $15 | 700M |
| Business | $39 | 1,800M |
Plans are paid from prepaid credit and renew only if you switch renewal on. Upgrade mid-month and you pay just the difference. Past the included credits, authorized requests bill at the usual $0.025 per million.
PLANS & USAGE
One number to compare.
Pay only for what the model actually used, or buy a monthly plan. Because credits track cost, you can compare any plan with pay as you go at a glance.
Create an account → verify your email → add $10 credit → set a spending cap.
Estimate your usage
A typical agent re-reads its context on every step, so most input is cached. Output is usually the smallest number and the most expensive per token.
PREPAID · NO SUBSCRIPTION
Estimated usage cost
Credits1,500M
Cached input$7.50
New input$25.00
Output$5.00
Add $10 or more; unused credit stays in your balance. This estimate does not include AGNT allowances and does not authorize a charge.
Add prepaid credit →DEVELOPER QUICKSTART
Start with a request.
One endpoint. The same key, balance and receipts.
HTTP API
Clear boundaries.
No surprises.
What is a credit?
One cached input token. New input tokens use 50 credits each and output tokens use 200, because they cost more to serve. Weighting the credit to cost is what lets one number compare a plan with pay as you go, whatever your agent sends.
Which model is this?
AGNT Flash: a one-million-token, tool-calling model with vision and JSON output, chosen and tuned for agent workloads. The name is stable; you will never need to change a model string when we improve what is behind it.
How do I pay?
Add $10, $25 or $50 through Stripe Checkout, or fund an agent with x402 USDC on Base. Authorized requests draw from prepaid credit. Monthly plans are optional and are paid from that same credit, so there is never an automatic card charge.
What happens if a request uses more than expected?
Before a request runs we reserve its worst case against your included credits and your stated cap. When it finishes we settle on the tokens it actually used and release the rest. You are never charged past the cap you authorized.
Are my prompts stored or used for training?
No. Prompts and completions are not retained after the response is delivered and are never used for training. Request receipts record token counts and charges only.
Give your agent a mind.
Long context, real tool calls, and a bill you can predict.
Already subscribe to AGNT?
Paid AGNT already includes monthly credits: Personal 100M, Always On 300M, Business and Enterprise 600M — per UTC month, no rollover. A Models plan can be bought on top, and the larger of the two allowances applies. AGNT can use this endpoint with no API key of your own.