By builders, for builders. Build on the trust layer for AI, created by @0G_Labs. Home of the latest private models, storage, DeFAI and decentralized agents.
No Gravity Zone
- Tweets 445
- Following 52
- Followers 2,184
- Likes 220
We had to pick a model for Ask AI on the 0G Builder Hub and the docs, so we ran the same question through five of them with the same 2000 token output cap.
Nothing here failed on quality. What separated them was where those tokens went.
1
4
178
glm-5.3 is the opposite. Deep thinking is always on, so with a 2000 token ceiling the reasoning used the whole budget.
Raise the cap or drop reasoning_effort and it answers fine. Our cap was tuned for a docs widget, not for a model that thinks this hard.
1
17
Router USD mode goes offline for two hours tomorrow, 02:00 to 04:00 UTC.
If your app can fall back to 0G mode, it will barely notice. Neither 0G mode nor Advanced Mode touches the payment layer, so both stay up.
Worth keeping a funded 0G balance around for exactly this reason.
3
4
240
Getting users funded on a new chain is usually the app developer's problem. You end up writing a bridging guide nobody reads.
This takes it off your plate. Bridge in from six chains, swap, and land on a page that points at apps to actually use.
0G Hub is live.
Bridge in from Ethereum, Solana, Base, Arbitrum, BNB Chain and Robinhood Chain.
Swap on 0G.
Find what to do with it, all in one place.
hub.0g.ai
5
17
466
🚨 Heads up if you use Router USD mode on 0G Private Computer.
We are upgrading the payment system behind it on Thu 24 Sep, 02:00 to 04:00 UTC.
USD mode is unavailable in that window and card top ups pause. 0G mode and Advanced Mode keep running, neither touches the payment layer.
1
4
8
355
Local times for the window:
10:00 to 12:00 SGT
04:00 to 06:00 CEST
03:00 to 05:00 BST
02:00 to 04:00 UTC
19:00 to 21:00 PDT on Wed 23 Sep
22:00 on Wed 23 Sep to 00:00 EDT
1
2
62
We rebuilt the 0G public RPC. Now geo-routed across 7 regions with HTTP/2 and WSS.
eth_call latency:
São Paulo 290ms → 24ms
Frankfurt 248ms → 26ms
Sydney 231ms → 34ms
London 252ms → 40ms
Singapore 76ms → 19ms
A typical dapp page load from Europe went from 2.4s to 0.3s.
8
29
17,497
Honest question for anyone building agents.
Would you let one spend its own money without asking you first?
44%
Already do
22%
Only with a hard cap
11%
Never, I hold the keys
22%
What could go wrong 😎
9 votes • Final results
3
151
Every new coding session starts with you re-explaining the project.
0G Memory stores the prompts, replies and tool calls encrypted on 0G Storage, so the next session already knows why you picked Postgres and what broke in auth last week.
github.com/0gfoundation/0g-m…
1
2
6
251
Cheapest model on 0G right now is Qwen-Flash: 2 cents per million input tokens, 1M context.
The priciest is o3-pro at $20 per million. Same API key, same balance, about a 900x spread.
Fun Fact: Most apps still send every call to one model.
pc.0g.ai
1
1
10
598
130 models on pc.0g.ai now. It was 64 last week, so someone has been busy.
GPT, Gemini and Claude are in there next to Qwen, GLM and DeepSeek.
Same key and same balance across all of them, which is the part we builders actually care about ;)
4
11
423
Friday. Three things you could ship on 0G before Monday:
💠 A RAG bot with embeddings and chat behind one API key.
💠 An app from a single prompt on app.0g.ai.
💠 An agent that leaves a signed TEE receipt per call.
Reply with what you shipped, we will take a look.
1
4
20
422
The detail worth noticing: single input up to 128,000 tokens.
Most embedding APIs stop at 8,192, so you end up chunking to fit the API instead of your data. Here you can chunk on meaning, or embed a whole doc as one vector.
Billed on input only, $0.0735 per 1M tokens.
Your RAG stack just lost a dependency.
Until today, 0G Private Computer covered chat, vision and code models. Embeddings were the missing call that kept a second provider around. Not anymore: same OpenAI-compatible API, same Router balance.
Embed and infer in one place.
1
4
14
384
Our intern is free tonight and has 0G App open 🤓
Drop a one line app idea in the replies. Best one gets built with a single prompt and posted back here.
People have been shipping landing pages, dashboards and wallet trackers this way since April.
3
4
18
422
Two trust modes on @0G_labs, and you pick per request.
Verified: cryptographic proof the real model served you.
🗝️ Private: the model runs sealed in a TEE, your prompts never leave the enclave.
One header. Same API either way.
Check it out at pc.0g.ai
GLM-5.3 got unexpectedly good at finding vulnerabilities 🤯
@Zai_org reports 84.5% on CyberGym, the best public score on that benchmark, and more than double its predecessor on exploitation. Which makes it the last model you want reading your source code.
On @0G_labs it now runs sealed inside a TEE, so you can point it at a private repo and neither we nor the provider can read what you sent.
Try it: pc.0g.ai/models/glm-5.3
1
1
13
413