Onchain inference, priced like a surplus. One API key → every LLM, paid in USDG on Robinhood Chain. 0xe1838f0dff03a9327dea87b2a2285850791297ec
Joined October 2026
- Tweets240
- Following49
- Followers1.6K
- Likes150
Surplus retweeted
Thank you @SurplusRH for including us in the ecosystem update 🙏
A bit more on what aethron-30b is, and what it's part of.
AethronAI is read-only onchain intelligence for Robinhood Chain, Base, BNB Chain, Arc and Solana. You paste a token or a wallet, and you get findings where every claim cites its evidence. No wallet connection, no keys, no buy or sell calls.
aethron-30b is the agent model behind it
A Qwen3-30B-A3B mixture-of-experts model (30.5B parameters, about 3.3B active), fine-tuned on real Aethron Agent sessions that passed our evidence validator. It calls read-only tools to scan tokens, trace wallets and explain transactions, answers only in short statements marked fact, inference or unknown, and replies in the language you pick. Weights are open on Hugging Face, Apache-2.0.
What we shipped this week
→ Wallet Watch: buy and sell alerts on wallets you follow, and alerts when several of them buy the same token
→ Notable wallets on every token: who received in the mint, from the creator, or first after launch
→ Token Scanner: top holders, most traded now, and flags for lookalike tickers
→ Alchemy-indexed data on all five mainnets
How the launch works
Every trade on Pons funds the model: 60% of creator fees pays for its GPU, 30% goes to the builder, and 10% buys and burns Surplus's token. At graduation, aethron-30b gets its own GPU on the Surplus API.
What's next
Testing aethron-30b on questions it has never seen, serving it inside Aethron Agent, and graduating on Surplus.
Read-only. Never asks for your keys. Not financial advice.
CA: 0xa694c90d513c17df0823d221eb7b1c94f8925fcf
🌐 aethronllm.xyz
🤗 huggingface.co/aethronai/aet…
Surplus retweeted
Lesson from going live: our LoRA got so used to calling tools that it stopped writing plans. Ask it for one and you get nothing back.
So the agent uses both. The fine-tune inspects the repo, the base model writes the plan. One vLLM server, base + LoRA in 14 GiB.
superqwenllm.com is now live on our own fine-tuned model.
Type a goal and gpt-oss-20b + our LoRA inspects a sample repo and drafts a plan for you to approve.
No sign-up. 5 tasks an hour. If the GPU is asleep, the first one takes 3–6 min.
Read-only until Sprint 3.
Surplus retweeted
We've aligned with @SurplusRH on $SUPERMUSE creator fees.
Dev takes 0%.
• 30% → $SUPERMUSE stakers, paid in ETH
• 60% → vault
• 10% → Surplus buyback
Claimed daily, split on-chain in one public tx.
Every fee goes back to the ecosystem.
Staking live today 👀
Surplus retweeted
Raeto is now revenue-generating.
Through our integration partners @pickhooks and @SurplusRH, AI requests are flowing through Raeto’s routing engine, and we collect a small fee on every single one.
This revenue doesn’t come from our own users. It comes from other apps choosing Raeto to decide which model handles each request. They save money, their users get faster and cheaper AI, and Raeto earns on every call.
Every new partner adds more requests. More requests mean more revenue. And that revenue is what powers the next stops on the line, including $RAETO staking, which shares real revenue with holders.
More to come.
Surplus retweeted
Raeto 🤝 @SurplusRH
A two-way partnership:
→ Raeto now serves Surplus models. Independent teams launch their AI models on Surplus, and you can pick them by name in Raeto chat and the API.
→ Surplus uses Raeto Auto routing to send each request to the right model.
Why it fits:
Surplus brings cheap, one-of-a-kind models. Raeto brings users and agents with AI credits to spend, plus Auto routing, which picks the right model for each request and saves Surplus up to 70% on AI requests.
Surplus retweeted
You asked for more demos, so here's one per tab: Fix, Refactor, Explain, Test.
Same model, four different jobs. Every answer is Artex's real reply, shown in full.
Try the tabs yourself: artexllm.com
Surplus retweeted
AethronAI: read-only onchain intelligence for anyone who wants to check a token or a wallet before trusting it.
What's inside:
Token Scanner
Paste any contract and get price, liquidity, 24h volume, market cap and holders, plus Aethron Sentinel: 11 rule-based checks, split in two.
→ Risks found: each flag shows the measured value, the threshold it crossed and the evidence behind it
→ Checks passed: what each pass shows, and what it doesn't prove
No single score. No "safe" badge.
Wallet Trace
Any address: balances in USD, transfers, counterparties and an interaction map.
Aethron Agent
Ask about a token, wallet or network in plain words. Every statement is labelled fact, inference or unknown and cites its evidence. Anything the evidence doesn't support is removed before you see it. Pick Surplus, Hugging Face models or aethron-20b, our own model.
Watchlists
Save tokens and follow how their signals change.
Robinhood Chain · Base · BNB Chain · Arc · Solana
Free to use. Read-only: it never asks for keys or signs anything.
Surplus retweeted
Charge AI graduated ⚡ 2 hours, 360+ buyers. Thanks @SurplusRH and everyone on the curve.
Next, by request:
→ Charge Wake: old launches waking up
→ direct Robinhood launches, then pump.fun
→ builder API, CHARGE unlocks premium
One token for every model. You pick the order 👇
Surplus retweeted
Charge AI is live on @SurplusRH ⚡
An open 30B model trained on 13,250 Pons launches. Paste a CA, get its read in a second.
The only real CHARGE is on Robinhood Chain:
0x7b8892cea949bc60d494681d1bc3ea96071ab0b8
Copies on Solana and Uniswap v4 aren't us.
chargeai.bid
Surplus retweeted
Our first sub-1-bit model is out.
Qwen3-8B at 0.55 bits per weight, built with LittleBit-2.
Early checkpoint: half an epoch of distillation on a single H200.
Open weights:
hf.co/LittleBitLLM/littlebit…
Surplus retweeted
252 downloads in 2 days.
Vela Agent Guard is a family of small guard models for AI agents that touch wallets and API keys.
The 0.6B one lost to plain regex on my synthetic eval. On real agent benchmarks it turned out to be the most balanced:
→ 6.7% false alarms on injection (4B: 19%)
→ 5.9% of legit tool calls blocked (4B: 32%)
→ 1.1% of bad tool calls let through
huggingface.co/VelaOrg/vela-…
Surplus retweeted
Up to 32B on the launchpad means our guard doesn't have to stay small.
Vela Guard 32B is in the works. Same three checks: leaked keys, injected tool outputs, risky tool calls.
Trained on messier, real-world-style data this time. Evals first, then it ships on Surplus.
Thanks @MCGlive and @ChillTRD for breaking down the thesis.
The short version: inference gets a market price, and open models get a launchpad that pays for their own GPUs.
What's live today:
→ 4 open models launched @Vela_LLM @LittleBit_llm @Artex_llm @useNockAI , all 4 graduated and listed on the Surplus API
→ 17.01M SURPLUS bought back and burned from their creator fees
→ one key for 450+ models, frontier and open
New today: the launchpad now takes models up to 32B parameters, served on 48 GB GPUs at $0.10 per 1M tokens.
More launches, more burn, more models listed.
Surplus retweeted
AI agents now hold keys, read the web and sign transactions. One hidden line in a tool output can empty a wallet.
$VELA is live on the @SurplusRH launchpad.
Vela Guard is an open guard model for agents that touch wallets and APIs:
→ finds leaked private keys, seed phrases and API keys
→ flags prompt injection hidden in web pages, token metadata, tx memos and email
→ rates every tool call allow / confirm / block before it runs
Live: model + code on Hugging Face, Apache-2.0
Next: hosted guard endpoint, SDK for agent frameworks
CA: 0x57390f221f8e8f14285db5df6cb30edcae75c2e8
Guard your agents.
Surplus retweeted
First score from the @googlegemma 4 Developer Agent Competition is in.
Nock Agent v1: 0.12. That's 12% of hidden coding tasks fixed, on one model, with a 4.5 minute cap per task.
Public notebooks sit between 0.06 and 0.18. We're in the middle. Next version changes one thing at a time.
Surplus retweeted
Closed source is dead.
Your agent's backend should be cheap too.
REA lets an agent take apart any app down to the binary and explain how a feature works.
- Native apps, Electron, websites, APKs, firmware, EVM bytecode
- Evidence for every claim
- Runs as an MCP server, analysis stays local
127K GitHub stars. 100% open-source.
github.com/morluto/rea
The LittleBit API now does tool calling, so agents like this can run on it.
- tools + tool_choice, up to 32 tools
- works with stream: true
- drop-in for any OpenAI SDK
$BIT stakers, Byte tier and up.
littlebitllm.com/stake
Surplus retweeted
The Nock Coder playground is live on nockai.dev.
Type any coding question in the box at the bottom. The answer comes from the real model, served on @SurplusRH.
No signup. Code first.
Surplus retweeted
$NOCKAI graduated!
In between:
- Nock Coder went live on Hugging Face.
- Benchmarks went public, misses and all.
- v2.1 beat its own base model on HumanEval.
- And now it's callable on @SurplusRH for $0.03 per 1M tokens.
Not slowing down.
$NOCKAI graduated. Congrats to @useNockAI \.
Third model off the Surplus launchpad, and already live on the API.
Nock Coder is a 1.5B coding model that answers with the code first. Call it with the same key as every other model:
model="launch/Nock-AI/nock-coder-1.5b@18859c0"
$0.03 per 1M tokens, served from the exact Hugging Face commit it launched with. Its creator fees fund the GPU, and 10% buys back and burns SURPLUS.
