Researchers Put a Brain Onchain

This is one of the most exciting developments I’ve seen this year.

There is now a large language model (LLM) blockchain contract.

That contract holds Qwen3.5’s 35 billion-parameter model.

Put differently, you can ask Qwen’s open-weight LLM a question and get a response, just like you would with OpenAI’s ChatGPT or Anthropic’s Claude.

The big difference: ChatGPT and Claude are under lock and key in a “walled garden.” Qwen’s LLM is open, giving it the ability to sit on a public blockchain.

In digital asset circles, it’s common to “check the chain.” That means you confirm a transaction took place.

With this latest development, we’ll need a new expression—“ask the chain.”

Users will send prompts and receive outputs the same way they send or receive cryptocurrency tokens.

This allows greater autonomy for prediction market outcomes, portfolio allocation tools, and even governance solutions. It’s a major step closer to The Permissionless Economy.

If it sounds like a cool project, that’s because it is. But it’s more than a fun experiment.

In fact, the implications are major.

And a big one is this: For the first time, a blockchain has a “brain.”

It Started as a Side Quest

The team at Gas Killer built the project. Ron Turetzky is one of the people behind it.

Turetzky is a developer who speaks at events related to decentralized autonomous organizations (DAOs). He’s aligned with collective ownership and views crypto as a road to “post-capitalism” where extractive corporations are a thing of the past.

It’s noteworthy that the team behind Gas Killer was focused on overcoming a gas constraint.

Public blockchains limit how much compute, or “gas,” can happen in each block. When we send typical transactions or conduct decentralized finance (DeFi) actions, this limit seems large.

But when it comes to recording a live tally of tens of thousands of votes onchain, gas limits become a roadblock. It’s a roadblock that Gas Killer is trying to remove.

It gained further legitimacy by running compute-intensive contracts such as the privacy-preserving DeFi solution RAILGUN.

This work led it to tackle a unique question…

What else can be done to support applications too compute-intensive for Ethereum or other Ethereum Virtual Machine (EVM) chains?

That gave rise to the idea of using an LLM as part of the infrastructure to aid governance.

LLMs can summarize proposals based on community inputs, aid deliberations, and even revive dormant DAOs. This line of thinking led the team to something unheard of, until now.

Recommended Links

“Market Wizard” Reveals Strategy to Help Americans Target $1,000

Larry Benedict, famous for handing his clients hundreds of millions of dollars in his career, is revealing his “two-rule” strategy to help his followers go for their first $1,000 in the days and weeks ahead. Join him for just $19.. Click here now.

Musk Opens "Rarest Wealth-Building Window Known To Man"

Jeff Brown calls it "the rarest wealth-building window known to man" – it's opened only four times in 160 years. Last time, Jeff used a special type of trade to rake in a mind-blowing 4,344% in a single year on a single stock... Now, he says Musk is opening it again. But it could slam shut on August 26. Click here now to watch Jeff’s urgent strategy session.

The $10 Million Prompt

The Gas Killer team translated the open-weight LLM Qwen from its current code base to Solidity.

Solidity is the main programming language for smart contracts that run on Ethereum and EVM-compatible blockchains. It’s the most popular code used in the public blockchain industry.

This meant all the LLM’s weights were translated and placed into a public blockchain contract.

Weights might seem like computer alchemy, but they’re just numbers stored in massive grids. This creates billions of multiplication and addition operations. These weights act like knobs that determine how a word or concept connects to the next. It’s how an LLM turns raw text input into intelligible outputs.

Gas Killer has two LLMs running on a blockchain—Qwen3.5-35B and the much smaller Qwen3-0.6B.

The team says the larger model uses about 3.6 trillion gas per answer.

That’s beyond massive.

To put that in perspective, the average limit on Ethereum is about 60 million gas units. That’s 60,000 blocks’ worth of compute for a large model, or about eight days of blocks… for one short answer. Obviously, that’s not realistic.

To get a sense of how difficult it would be to host this model on Ethereum, consider this: Paying for that much gas per prompt would cost around 324 ETH, or roughly $616,000.

And gas is very inexpensive right now. If prices rose to normal levels, this prompt would quickly top $10 million.

We can see why solutions such as Gas Killer are needed. The compute is just too intense to run on a blockchain.

Which is why it runs the LLM contract on an offchain fork of Ethereum. It’s how the compute becomes manageable.

Users can still send an onchain transaction that queries the LLM contract via a specific interface. The contract runs and returns the output with a simple hash saying who ran the solution. There’s more to the madness, but it’s outside the scope of today’s essay.

What we need to know is that this solution lets us send a prompt the same way we’d send a typical transaction onchain. We then receive an intelligible output.

Which is where things get exciting…

Where This is Going

Longtime readers might recall a discussion on Polymarket and its oracle solution UMA called The Polymarket Competitor Nobody Is Talking About. The story discussed a dispute over whether Ukrainian President Volodymyr Zelenskyy wore a suit.

More importantly, the discussion highlighted the need for impartial oracles to give straightforward responses to event-driven contracts and markets.

Put simply, there must be a fair and impartial way to determine whether an event happened. What Gas Killer created could easily be that solution.

There’s also portfolio management… Increasingly, LLMs are advising (and sometimes deciding) how we manage our capital in our brokerage account, cryptocurrency wallet, or even daily spending.

An onchain LLM could rebalance a portfolio every day. Or hedge one in response to a global event.

These are two low-hanging fruit outcomes.

Of course, questions remain about how these solutions will come to market. The prototype from Gas Killer is very new. It still is a bit spotty when you give it a try, and there are kinks to work out to bring it to a more production-friendly state. That’s expected for such a novel concept.

The bigger piece of the puzzle will be the need for cryptoeconomic solutions that ensure the provider is operating as it should, as trustless and verifiable LLMs.

This, in turn, will result in more demand for blockchains and even give solutions like Liquid Restaking Tokens greater product-market fit as they can supply the cryptoeconomic security. That’s great for asset prices and various other assets. It should also attract developers and bring novel solutions to market.

We’re in the early innings of a future where greater autonomy happens day-to-day, onchain.

It’s truly a machine-to-machine economy that will run without censorship on permissionless rails. In time, it will make centralized models seem outdated.

This is possibly the biggest leap for the ecosystem this year.

Your Pulse on Crypto,

Ben Lilly
Editor Chain of Thought