Builders update
The Builders Update.
The changelog for everyone building on 0G: network updates, new features across the stack, events, builder spotlights, and the AI news that matters. One block at a time, every week or two.
Back from Seoul and Singapore
Block #03
Block #03 covers Sep 28 to Oct 12. The team spent it between KBW in Seoul and TOKEN2049 week in Singapore, and product kept shipping anyway.
Encrypted Transit: the Router never sees your prompt
Requests are now encrypted at the TEE Gateway and stay encrypted through the Router. They're only decrypted inside the TEE broker before they reach the provider. On models marked Private, decryption and inference both happen inside the enclave, so the whole path is end-to-end encrypted. It's live for GLM-5.3 (all providers), 0gm-1.0-35b-a3b and both Qwen3 Embedding models, and the model list now labels every model Private or Verified.
New on the Router: Opus 5.5, decision models and private embeddings
claude-opus-5.5 is live with 1M context, tool calling and prompt caching at $4 / $20 per 1M tokens. A new Decisions tab lists models that return structured answers instead of text, built for routing, triage and guardrails: vela-2.0-9b ($0.04 per 1M input, runs in a TEE) and Laya (421M, Private). Qwen3 Embedding 8B ($0.01/M) and 0.6B ($0.005/M) now run Private, and 21 Flash models got a second serving route with lower time-to-first-token at the same price.
Also: Seedream 4.5, Seedream 5.0 Lite and minimax-image-01 for images. Seedance 2.0/2.5 and MiniMax-H3 video now support first/last-frame control and multimodal references, and there's a new GET /v1/videos/{id}/artifacts endpoint (beta).
Your inference budget can now come from staking
Compute Finance is live. Stake 0G on Ascend to get a0G, then use it to mint Infinite AI (iAI). Eligible staked iAI is designed to receive 1.271 compute credits a day, refreshed at 00:00 UTC and spendable across 100+ models on Private Computer and the 0G App. Under initial parameters, that's about 1,030 agent actions or 62 code reviews a day per iAI.
The launch moved fast: 667 wallets minted 3,800+ iAI in the first seven hours, and Ascend passed $4M TVL within a week. Comfy, the ecosystem's zero-gas token launchpad, saw 170+ launches in its first 24 hours, with agent launches coming soon.
iAI and its credits are not available in all jurisdictions and are subject to eligibility and product terms. Credits are only for supported 0G AI services, are not redeemable for cash, and are not interest or a guaranteed return. Not financial advice.
0G Dev Day: Zero to Infinity
On Oct 5, Dev Day took over the National Gallery Singapore with 2,700+ registrations and Senior Minister of State Tan Kiat How as Guest of Honour. 0G signed an MoU with NTU, announced Apollo Accelerator Cohort 2, and committed S$20m over five years as Singapore becomes its global headquarters. Between talks, people trained a robot dog, shipped builds in the Hacker Zone and collected Passport stamps for a 0G Labubu.
The Hacker Zone launchpad works at home too
The Dev Day build stations ran Claude Code and OpenCode on 0G models with all 25 skills preinstalled. Every build idea was a copy-paste prompt, from a 10-minute chatbot to an onchain guestbook or an agent with its own Agentic ID. The page stays up, so run the installer, get an API key from Private Computer, and pick an idea.
Seoul, then TOKEN2049
At KBW, Michael took the 0G Stage to talk about the trillion-dollar agentic economy. He also told CoinDesk Live that if an agent is spending your money, you should be able to verify what it ran. House of AI Seoul with American Fortress drew 680+ registrations, and at TOKEN2049 the 0G booth hosted Bond, Zia and whale.dance alongside the team.
Bond is building one financial super app on 0G
Bond's answer to money spread across too many chains, apps and vaults is a single super app built on 0G, now in public beta. CEO Taweh Beysolow made the case on the KBW Stable Stage and again at Dev Day, and the team ran the 0G booth with us at TOKEN2049.
Your coding agent can build on 0G now
Block #02
Block #02 covers Sep 3 to Sep 28. The team was heads-down in Singapore getting ready for Dev Day, and a lot shipped anyway.
One line sets your coding agent up for 0G
Run curl -fsSL https://build.0g.ai/install | sh in your project. It asks which assistant you use and which parts of 0G you're building on, shows you the plan, and waits for your go-ahead. It then clones 25 agent skills into ./.0g-skills (Storage, Compute, Chain, Agentic ID, plus Private Computer setup for Claude Code) and prints the SDK install line and .env keys you need. It never touches your CLAUDE.md, rules files or settings, and we rebuilt the Builder Hub homepage around it.
130 models behind one API key
The Private Computer catalog went from 64 to 130 models in a week. GPT, Gemini and Claude are now available alongside Qwen, GLM, DeepSeek and Baidu's ERNIE family, all on the same key and balance. Embeddings landed too: qwen3.7-text-embedding takes a single input of up to 128K tokens and only bills input, at $0.0735 per 1M, so your RAG stack doesn't need a second provider.
GLM-5.3 now runs sealed in TeeML
Z.ai's latest flagship took over the Private tier from GLM-5.2. It runs fully inside a TEE enclave where neither 0G nor the provider can read your prompts. If you were pinned to glm-5.2 in Private mode, migrating is one line: change the model name to glm-5.3.
The public RPC got a lot closer to you
We rebuilt the 0G public RPC, and it's now geo-routed across 7 regions with HTTP/2 and WSS. eth_call latency dropped from 290ms to 24ms in São Paulo, 248ms to 26ms in Frankfurt and 231ms to 34ms in Sydney. A typical dapp page load from Europe went from 2.4s to 0.3s.
Your users can bridge in from six chains
0G Hub went live on Sep 22. Users can bridge in from Ethereum, Solana, Base, Arbitrum, BNB Chain and Robinhood Chain, swap on 0G, and find apps to use on the Discover page. If you've been writing your own "how to get funds onto 0G" guide, you can send users here instead.
Two demo days, 24 teams
Taipei fit a full hackathon into one day at Tempo House: morning workshops, a 3-hour build sprint and 13 working demos on stage by evening. A week later, 11 teams from the 0G Atlas Founder House cohort, run with HackQuest, demoed agent-economy builds live, from trading agents and skill markets to payment rails and agent memory.
Pulsar Money and Ask Max run their AI on 0G
Pulsar Money runs its in-app financial models on 0G Private Computer, so each response carries proof of which model ran and where. On whale.dance, Ask Max stores every conversation on 0G Storage as the user's own asset and runs inference on 0G Compute, and 0G keeps no prompts or responses.
Hello, builders
You're reading block #01
Every week or two, everything that matters to 0G builders gets sealed into a block on this page: protocol news, fresh ships, events, builder spotlights, and the occasional AI headline worth your attention. Append-only, no reorgs, history is history. And yes, this page shipping is itself news, so the genesis block includes itself.
The Testing Program is live
Break 0G things responsibly and get paid for it: sign up with a wallet, hunt bugs across 17 targets (apps, infra, ecosystem) and climb L0 → L3 for up to 100 0G in Compute Credit.
Ask AI answers your 0G questions
The hub now speaks: /ask runs GLM-5 inference on 0G Compute and answers questions about building on 0G in plain language. We build on the thing we tell you to build on.
0G Japan: 日本のビルダーへ
A home base for the Japanese community: events, resources, and localized onboarding, live now on the hub.
All quiet on the mainnet
When the protocol moves (upgrades, planned releases, forks), you'll read it here first, translated into builder impact. This week the chain just hums. That's a feature.