OpenGradient ships verifiable inference

OpenGradient has gone live on Virtuals Protocol, bringing a system that lets developers prove an artificial intelligence model actually produced a given output. (app.virtuals.io) The company says its network has already processed more than 2 million verified inferences and generated more than 500,000 cryptographic proofs. OpenGradient said those model runs are handled across a decentralized network of Trusted Execution Environment and graphics processing unit nodes. (coingabbar.com) (docs.opengradient.ai) In plain terms, an inference is a model call — the step where an artificial intelligence system turns a prompt into an answer. OpenGradient’s docs say it routes large language model requests through hardware-secured environments that can attest which prompt and computation were used. (docs.opengradient.ai) The company separates the fast step, where the model answers, from the settlement step, where proof is checked and recorded on its ledger. OpenGradient says that design is meant to keep response times closer to standard cloud services while still making runs auditable by the network. (docs.opengradient.ai 1) (docs.opengradient.ai 2) That pitch fits Virtuals Protocol’s core business: a marketplace and tokenization system for autonomous agents that buy services, complete jobs, and move money onchain. Virtuals says its platform is built around artificial intelligence agents acting as economic actors and completing commerce with humans and other agents. (virtuals.io) (whitepaper.virtuals.io) For those systems, the missing piece has been trust in the model call itself. OpenGradient’s materials argue that most users can see an agent’s action onchain but cannot verify which model ran, what input it received, or whether the output path was altered. (opengradient.ai) (docs.opengradient.ai) OpenGradient is also trying to tie that verification to payments and developer tooling. Its software development kit uses x402, a payment-gated Hypertext Transfer Protocol standard built around the 402 Payment Required status code, to let applications pay for verifiable large language model inference by request. (docs.opengradient.ai) The company disclosed $9.5 million in total funding this week, led by Andreessen Horowitz’s crypto arm, with Coinbase Ventures and SV Angel also named as backers. OpenGradient said the money will be used to scale a network that hosts, executes, and verifies model inference. (aithority.com) Virtuals’ launch page describes OpenGradient as infrastructure for “verifiable on-chain computing and AI inference,” with the token used to access compute, secure the network, and reward participants. The immediate test is whether developers building agents on Virtuals use proof-backed model calls in production rather than as a demo feature. (app.virtuals.io)