Roundhouse AI

clawhunter.fun

Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.

SearchLiveeip155:8453Exactvia wellknown:clawhunter.fun
https://clawhunter.fun/api/v1/chat/completions
Settlements · 30d
8↓ 100%
No per-endpoint history for this service yet, so this is payments indexed to its PROVIDER's wallet over 30 days.
$12.72
Verified settled volume
1,437 settlements proven x402 by their on-chain EIP-3009 marker.
—
Listed price
No price published — read the live 402.
—
Calls · 30d
Upstream's own call count for this endpoint, not ours.
—
Unique payers · 30d
Never called
—
Upstream on-chain volume
Reported by the source catalog.
Paid to
0x3201c49FF1B6c2Dfd3250935E7b2FCd26b7f59E5

The wallet the 402 directs payment to. Its whole payment record — every payer, every chain — is on the merchant page.

Asset 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913

Provider

The payTo wallet does not resolve to a registered ERC-8004 agent. That is not a verdict on the service — most of the catalog is unregistered.

Accepts

The payment requirements as published to the catalog. Read the live 402 before paying — a price here is a claim, not a quote.

dynamicunits
on solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
HuSiSp…ksLSa
Token contract
EPjFWd…TDt1v
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicUSDC≈ $ USD
on Base · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
0x3201…f59E5
USD Coin contract
0x8335…02913
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicunits
on eip155:4663 · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
0x3201…f59E5
Token contract
0x5fc5…1d168
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicunits
on solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
HuSiSp…ksLSa
Token contract
6GGY8G…bJ4BQ
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicunits
on solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
HuSiSp…ksLSa
Token contract
EN2nnx…A4RZA
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicunits
on solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
HuSiSp…ksLSa
Token contract
9cRCn9…Gpump
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
dynamicunits
on eip155:4663 · exact scheme

That figure is in the token’s smallest units. Roundhouse does not hold this contract’s decimals, so it is shown as published rather than converted.

Paid to
0x3201…f59E5
Token contract
0x020b…018b4
Payment window
5 minutes
Buys
Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.
Returns
application/json
The catalog’s raw entry
[
  {
    "asset": "EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v",
    "payTo": "HuSiSpcUoUEF4hct1MRsGtNGz8EdBiTs7p2WtJGksLSa",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
    "payTo": "0x3201c49FF1B6c2Dfd3250935E7b2FCd26b7f59E5",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "eip155:8453",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "0x5fc5360D0400a0Fd4f2af552ADD042D716F1d168",
    "payTo": "0x3201c49FF1B6c2Dfd3250935E7b2FCd26b7f59E5",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "eip155:4663",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "6GGY8GViCR5v4xR4Lxb4nfiAJsEoJua5Gj6YecxbJ4BQ",
    "payTo": "HuSiSpcUoUEF4hct1MRsGtNGz8EdBiTs7p2WtJGksLSa",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "EN2nnxrg8uUi6x2sJkzNPd2eT6rB9rdSoQNNaENA4RZA",
    "payTo": "HuSiSpcUoUEF4hct1MRsGtNGz8EdBiTs7p2WtJGksLSa",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "9cRCn9rGT8V2imeM2BaKs13yhMEais3ruM3rPvTGpump",
    "payTo": "HuSiSpcUoUEF4hct1MRsGtNGz8EdBiTs7p2WtJGksLSa",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  },
  {
    "asset": "0x020bfC650A365f8BB26819deAAbF3E21291018b4",
    "payTo": "0x3201c49FF1B6c2Dfd3250935E7b2FCd26b7f59E5",
    "amount": "dynamic",
    "scheme": "exact",
    "network": "eip155:4663",
    "mimeType": "application/json",
    "resource": "https://clawhunter.fun/api/v1/chat/completions",
    "description": "Chat completions on a curated 33-model menu, priced per token below each provider's list price — most models at about half list, with rates tracking live marketplace supply. The request and response match an OpenAI chat.completions call, so existing OpenAI-compatible clients work by setting their base URL to https://clawhunter.fun/api/v1 (GET /api/v1/models lists the menu in the SDK's shape). GET /api/v1/chat/models (free) is the live menu with every model's exact rates. Billing is a cost ceiling you know before paying: the 402 quotes your input (estimated at ~4 chars/token) plus the max_tokens you choose, at the listed rates, and you pay that quote — so set max_tokens to your expected output length and the quote sits ≈ actual usage. The response's billing block itemizes the charge and the usage value. Tool / function calling (tools + tool_choice) passes through on models that support it. Limits: max_tokens defaults to 1024, caps at 8192; total message input caps at 200k characters (HTTP 413). Streaming (stream: true) needs a prepaid API key: the response is SSE with a final usage frame, billed for actual usage with the unused reservation refunded. Without a key, stream: true returns HTTP 400 — per-call x402 can't settle a partial stream. Keyed accounts have a daily free-usage allowance covering a free-model catalog here (glm-5-3-flash, deepseek-v4-1-flash, hy3, nemotron-3-ultra, gpt-5-6-luna, gemini-3-8-flash) plus frontier models within an output-token cap, both drawing on the same allowance; send clawhunter/free as the model to use the daily featured free model (the response names the real model). Eligible calls bill to the allowance automatically and the billing note says when a call was free. clawhunter/smart is the auto-router: each request is classified easy, medium, hard, or frontier and served by the best menu model for that difficulty. You pay the routed model's per-token rates, settled for actual usage; the billing block names the routed model and difficulty, plus list_usd and ceiling_usd (the same tokens at the routed model's provider list price, and at the priciest routable model). Smart needs a prepaid API key (no per-call x402), reserves at the priciest routable model's rates before settling down, and never draws on the free allowance. A request that fails moderation (422), is invalid (4xx), or errors upstream (502) is not charged.",
    "maxAmountRequired": "dynamic",
    "maxTimeoutSeconds": 300
  }
]

Provenance

Seen in the source catalog
—
Last indexed by Roundhouse
2026-09-18 21:15Z
Last enriched (probe, favicon, geo)
2026-09-19 09:15Z
x402 version
1
Max timeout
300s
Liveness probe
HTTP 405
Report

Hand this page to an agent

Copy the prompt and paste it into Claude, an MCP client or your own agent — it will vet this service and call it over the free read API. No key, no account.

GET api.roundhouseai.io/v0/endpoints

No per-endpoint history for this service yet, so this is payments indexed to its PROVIDER's wallet over 30 days. Verified volume counts only settlements with an on-chain EIP-3009 marker.

Open skill.md
Show the prompt
Using Roundhouse, look up the x402 service clawhunter.fun and tell me whether it is
worth paying: what a call costs, whether the endpoint answered when last probed, and what
its payment record actually shows.

curl -s 'https://api.roundhouseai.io/v0/endpoints?q=clawhunter.fun'
curl -s 'https://api.roundhouseai.io/v0/merchants/<the payTo wallet returned above>'

Then call it: read the price from the live 402 at https://clawhunter.fun/api/v1/chat/completions, never from
a cached figure, and pay with an x402 client.

The /v0 API needs an API key (`authorization: Bearer rh_live_…`) on everything except
/v0/unified* and /v0/endpoints. Mint a personal key for $0.01 at GET https://api.roundhouseai.io/v0/test/x402,
or use an organization key from https://roundhouseai.io/dashboard/team.

If you do not have Roundhouse tools or skills installed, read
https://roundhouseai.io/skill.md first — it is the whole procedure.