0G Labs

**0G Onsite Lagos** wrapped its first day with 50 builders working on robotics-focused projects using real hardware—not simulations. **Key developments:** - Verifiable inference implementation - Persistent agent memory systems - Trustless coordination protocols - Deployment to actual humanoid robots The two-day event marks 0G's third return to Lagos, with live demos scheduled for day two. Builders are integrating 0G's infrastructure directly into physical robotics applications.

0G Labs published a comprehensive recap of their participation at Consensus Miami 2026. The team maintained an active presence throughout the week-long conference with: - **5 hosted events** bringing together community members and partners - **5 speaking slots** across different stages - **2 community activations** engaging attendees - **1 keynote presentation** focused on the Trillion-Dollar Agentic Economy The full details of their activities and insights from the conference are available in their [official blog post](https://0g.ai/blog/consensus-miami-2026-recap). This marks a significant showing for the DeAI L1 ecosystem at one of the industry's premier conferences.

At Consensus Miami's Agentic Tooling Summit, Michael H outlined the infrastructure gaps preventing AI agents from scaling into a trillion-dollar economy. **Three critical requirements identified:** - **Verifiable compute** - ensuring agent actions can be cryptographically proven - **Sovereign agent identity** - establishing autonomous identity standards (Agentic ID, ERC-7857) - **Onchain coordination rails** - enabling agents to interact and transact seamlessly The keynote emphasized that while demos show promise, production-ready agentic systems need these foundational layers before autonomous economies can emerge at scale.

**ERC-7857 introduces onchain identity verification for AI agents** A new standard called Agentic ID provides AI agents with blockchain-based identification and verification. The system stores ownership records and attestations directly onchain. **Key features:** - Agent state data encrypted on 0G Storage - Automatic re-encryption with each ownership transfer - Verifiable proof of agent identity, ownership, and model execution The standard enables users to confirm which specific agent performed an action, verify current ownership, and validate that the correct model executed as intended.

**0G Labs successfully trained a 107-billion parameter AI model** using its DiLoCoX framework in partnership with China Mobile in 2025. **Key achievements:** - 357x more communication-efficient than traditional methods - Largest documented decentralized AI system (48% larger than Bittensor's models) - Runs on standard 1 Gbps internet connections without data centers **Future roadmap:** - Scale to 700B+ parameters - Implement 100M token context window This marks a shift from theoretical concepts to practical implementation of decentralized AI training at scale.

**0G Storage** has launched a decentralized storage solution specifically designed for AI workloads rather than archival purposes. **Key features:** - Up to **90% cheaper** than AWS S3 - **2 GB/s throughput** with GPU-accelerated erasure coding - **Onchain provenance** for every data blob - **Mutable, real-time access** - a capability competitors like Arweave and Filecoin lack The platform positions itself as filling the "hot storage" gap in the decentralized storage market, offering infrastructure for training data, model weights, and agent memory that users can verify themselves. Live statistics available at [storagescan.0g.ai](http://storagescan.0g.ai)

0G has launched **Private Computer**, a system that runs every AI inference request inside a Trusted Execution Environment (TEE). Key features: - Hardware-verified privacy for all AI operations - No third parties can access data—not even the service provider - Privacy built into the runtime, not just a cloud feature checkbox The platform aims to eliminate trust requirements when using AI services. Users can experience the system at [pc.0g.ai](http://pc.0g.ai). This represents a shift from privacy as an optional feature to privacy as fundamental infrastructure for AI applications.

**0G has released 0GM-1.0-35B-A3B**, their first proprietary AI model trained entirely on their decentralized GPU network. **Key specifications:** - 35B parameter Mixture of Experts (MoE) architecture with 3B active parameters - 262K token context window, extensible to 1M tokens - Apache 2.0 open-source license - Served inside a Trusted Execution Environment (TEE) on [pc.0g.ai](http://pc.0g.ai) **Performance highlights:** - Outperforms Qwen 3.6 35B on 12 of 14 MMLU-Pro subjects - Achieves 83.3% on AIME 2026 (+13.3 percentage points improvement) The model represents a shift from hosting AI to building sovereign AI infrastructure, with every inference TEE-attested for verifiability. Weights are available on [HuggingFace](https://huggingface.co/0G-AI/0GM-1.0-35B-A3B-0427).

**DSRV Labs published an independent analysis of 0G Private Computer**, highlighting a fundamental choice in AI privacy: relying on policy promises versus code-enforced guarantees. **Key technical approach:** - Every AI inference request runs inside a Trusted Execution Environment (TEE) - Hardware-level attestation proves what operations occurred - Zero visibility for anyone in the supply chain, including providers - Sealed inference architecture ensures end-to-end privacy The external review validates 0G's technical implementation of privacy-by-design rather than privacy-by-policy.

**0G Pay** has launched a streamlined payment solution that reduces the traditional four-step funding process to a single swipe. The new system allows users to: - Top up their account directly with a card - Spend $0G tokens onchain immediately - Eliminate multiple transaction steps This follows a broader trend in blockchain UX improvements, similar to Polygon Trails' one-click money transfer solution launched earlier this year. The update addresses a key friction point in crypto adoption by simplifying the onchain payment experience.