# Prism Network Batch LLM Inference (GPU Rental, Onchain Settlement)

> Prism Network Batch LLM Inference (GPU Rental, Onchain Settlement) is a paid API for AI agents from api.prismnetwork.tech, paid per call via x402, $0.012144/call, status unknown (last checked 2026-10-11).

Runs multiple LLM prompts in a single batch on rented NVIDIA GPU compute, billed per-second in USDG and settled onchain with a Merkle-verified receipt.

## Facts

- Endpoint: POST https://api.prismnetwork.tech/inference/v1/batch?utm_source=zero.xyz
- Price: $0.012144/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-11
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/prism-network-batch-llm-inference-gpu-rental-onchain-settlement-58a4f2b6
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_wJJOWlJsc8KU-oGv7MGJ4

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability prism-network-batch-llm-inference-gpu-rental-onchain-settlement-58a4f2b6 -d '<json body>'
```

Example prompt: Run these 3 prompts through llama3.1:8b on Prism — 'What is metered GPU compute?', 'How does onchain settlement work?', 'What is USDG?' — cap each response at 200 tokens and give me a receipt.

## When to prefer this

Choose this endpoint when you need to process multiple independent prompts in one API call and require a cryptographically verifiable, onchain receipt of the compute spend — ideal for AI agents that must audit their own inference costs, compliance workflows that need tamper-evident proof of LLM outputs, or any pipeline where you want to minimize round-trips while paying trustlessly from a wallet. Prefer over single-inference endpoints when batch size is 2–64 prompts.

## Known failure modes

- Insufficient wallet balance or payment failure — returns 402 Payment Required
- Model name not in enum ('llama3.2:3b' or 'llama3.1:8b') — validation error
- Prompt array empty or exceeds 64 items — returns schema validation error
- Individual prompt exceeds 32 KiB — item rejected or truncated
- num_predict out of range (must be 1–1024) — validation error
- GPU lease unavailable — service temporarily unavailable
- Onchain settlement failure — receipt may not include settlement_tx

## How this service works

GPU compute your agents can rent with a wallet. Lease NVIDIA capacity, pay per second in USDG, and settle every lease onchain with a public receipt.

## Output

Returns a JSON object with: an array of per-prompt items each containing the generated response text, token usage (prompt + completion tokens, duration in ms), a lease ID, a sha256 commitment of the prompt and response, and a Merkle proof; plus aggregate token usage across the batch and a signed onchain receipt with the payer address, Merkle root, list of lease IDs, amount paid in micros, and the settlement transaction hash.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "model": {
   "enum": [
    "llama3.2:3b",
    "llama3.1:8b"
   ],
   "type": "string",
   "description": "Which model answers every prompt."
  },
  "options": {
   "type": "object",
   "properties": {
    "num_predict": {
     "type": "integer",
     "maximum": 1024,
     "minimum": 1,
     "description": "Maximum output tokens per prompt. Lower is cheaper."
    }
   },
   "description": "Generation options, applied to every prompt. The price scales with the output cap."
  },
  "prompts": {
   "type": "array",
   "items": {
    "type": "string",
    "minLength": 1,
    "description": "One prompt, up to 32 KiB."
   },
   "maxItems": 64,
   "minItems": 1,
   "description": "Independent prompts, answered in the order given."
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "count": 2,
  "items": [
   {
    "index": 0,
    "usage": {
     "duration_ms": 640,
     "prompt_tokens": 12,
     "completion_tokens": 14
    },
    "lease_id": 1047,
    "response": "Metered GPU compute bills by the second and settles onchain.",
    "commitment": {
     "index": 0,
     "model": "llama3.2:3b",
     "prompt": "sha256:9f2c...",
     "lease_id": 1047,
     "response": "sha256:41ab...",
     "prompt_tokens": 12,
     "completion_tokens": 14
    },
    "merkle_proof": [
     {
      "hash": "sha256:7d10...",
      "side": "right"
     }
    ]
   }
  ],
  "model": "llama3.2:3b",
  "usage": {
   "prompt_tokens": 25,
   "completion_tokens": 31
  },
  "receipt": {
   "count": 2,
   "model": "llama3.2:3b",
   "payer": "0x0000000000000000000000000000000000000001",
   "version": 1,
   "algorithm": "rfc6962-sha256",
   "issued_at": "2026-08-24T09:00:00.000Z",
   "lease_ids": [
    1047,
    1048
   ],
   "merkle_root": "sha256:c3e8...",
   "paid_micros": "12144",
   "settlement_tx": "0x..."
  }
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/prism-network-batch-llm-inference-gpu-rental-onchain-settlement-58a4f2b6/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from api.prismnetwork.tech](https://www.zero.xyz/host/api.prismnetwork.tech/llms.txt)