AdFreeLLM
Live paid inference

Eight models.
Pay for tokens,
not seats.

OpenAI-compatible inference for agents, automations, and developers. Choose a simple flat rate through Halo or model-specific pricing through AntSeed. Settle in USDC on Base.

Choose a payment rail Read catalog.json
8 modelsReasoning, code, chat
2 payment railsHalo and AntSeed
Base + USDCMachine-native settlement
OpenAI protocolFamiliar request shape
Two ways to buy

Pick the billing model that fits the workload.

Both routes reach the same eight-model fleet. Halo is simplest for mixed workloads; AntSeed gives price-sensitive agents model-level control.

Flat token rateBest for mixed workloads

Halo

1 USDC / million total tokens

One predictable rate across all eight models. Discover the operator through Halo, fund a buyer Vault, and route requests without managing a separate vendor subscription.

  • Eight models at the same published token rate
  • Public operator discovery and availability
  • Base USDC settlement through the Halo Vault rail
Browse Halo models Verify models
Per-model rateLowest entry price

AntSeed

0.0414 USDC / million input tokens, from

Choose a model-specific input and output rate. The provider publishes signed metadata, capabilities, load, pricing, and OpenAI-compatible protocol support for machine discovery.

  • Signed public provider metadata with eight services
  • Separate input, output, and cached-input prices
  • Tool use and structured output declared for every model
Buy with AntSeed Live metadata
Model fleet

One catalog for reasoning, code, speed, and Chinese.

Every advertised model supports the OpenAI Chat Completions request shape. Capabilities below are taken from the live provider metadata.

claude-opus-5

High-end reasoning and coding workloads.

reasoningtoolsstructured

claude-sonnet-5

Balanced agent, chat, and code work.

reasoningtoolsstructured

deepseek-v4-flash

Fast coding, chat, and math requests.

fasttoolsstructured

deepseek-v4-pro

Reasoning-heavy code and math work.

reasoningtoolsstructured

glm-5.2

Reasoning and Chinese-language tasks.

chinesetoolsstructured

kimi-k2.7-code

Code generation and mathematical work.

codingtoolsstructured

minimax-m3

Fast chat and lightweight automation.

fasttoolsstructured

qwen3.5-397b-a17b

General reasoning, code, and agent work.

reasoningtoolsstructured
Transparent AntSeed rates

Model-specific launch pricing.

USDC per one million tokens. Halo remains a flat 1 USDC per one million total tokens; compare against these AntSeed input/output rates for your traffic mix.

Current acquisition pricing · published by signed provider metadataCached input is 10% of input rate
ModelInput / 1MOutput / 1MCached input / 1M
claude-opus-51.055.250.105
claude-sonnet-50.362.400.036
deepseek-v4-flash0.04140.08250.00414
deepseek-v4-pro0.15840.31680.01584
glm-5.20.421.320.042
kimi-k2.7-code0.133680.62370.01338
minimax-m30.0840.3360.0084
qwen3.5-397b-a17b0.1440.8640.0144
Built for discovery

Agents should not have to read a sales page.

The machine catalog exposes models, payment rails, rates, capabilities, and live verification endpoints in JSON. Use the source directories when fresher network state matters.

/llm-services/catalog.jsonlocal catalog → relay.runhalo.xyz/v1/modelsHalo → 1.14.137.105/metadataAntSeed →
agent discovery
// Start here
GET https://adfreellm.com/llm-services/catalog.json

// Then select a rail
{
  "halo": {
    "price_usdc_per_million_total_tokens": 1,
    "discovery": "https://relay.runhalo.xyz/v1/models"
  },
  "antseed": {
    "pricing": "per-model input/output",
    "metadata": "http://1.14.137.105/metadata"
  }
}
No monthly plan

Start with the rail your agent already understands.

Use Halo for one flat token price, or AntSeed when per-model economics matter. Both expose live public discovery before you commit funds.

Browse Halo modelsAntSeed guide