Agents & Automation / Somnia LLM Inference

Somnia LLM Inference

Live

Deterministic LLM inference, powered by the Somnia Validator Network. Every validator runs the same pinned model and must return a byte-identical answer.

Product page Docs

Requests 24h4,218

Success99.6%

Median latency5.8s

Validators / requestUp to 10Set by the calling contract

Pinned model

Qwen3-30B

Serving

Weights: Open, pinned hash
Temperature: 0
Seed: Fixed
Output: Byte-identical
Agent id: 1284729384…
Price / call: 0.0042 SOMI

Output modes

Four ways to ask

Constrained text	Pick one of a fixed set of labels	classify("…", ["YES", "NO"])
Bounded numbers	Return a number inside a range	score("…", 0, 100)
Chat	Free-form completion, byte-identical across validators	chat([{ role: "user", … }])
Tool calls	Structured calls the contract can execute	tools([getPrice, …])

Latest

Inference requests

Request	Caller	Mode	Agreement	Tokens	Latency	Answer
3919402	OracleCore	Constrained text	3 / 3	412	4.9s	France
3919388	ModerationRegistry	Constrained text	3 / 3	236	3.1s	APPROVED
3919377	0x9e14…b7a0	Bounded numbers	5 / 5	690	6.4s	72
3919350	NameRegistry	Chat	3 / 3	158	2.7s	"Bitcoin 20-min up"
3919331	0x41b0…9a07	Tool calls	3 / 3	1,022	8.8s	getPrice(BTC)

Example

Call the model from a contract

solidity

uint256 id = somniaAgents.requestInference{value: fee}(
    InferenceRequest({
        model: "qwen3-30b",
        mode: Mode.ConstrainedText,
        prompt: "Who won France vs Italy on 2 Oct 2026? Answer with one label.",
        labels: ["France", "Italy", "Draw"],
        validators: 5              // up to 10, set by the caller
    })
);

function onInference(uint256 id, bytes calldata answer, Receipt[] calldata r)
    external onlyAgents { ... }