Last month, someone mined the first quantum-safe Bitcoin transaction on the mainnet, using a construction developed by StarkWare that protects coins without requiring the whole network to change its software. The catch was the cost: building that transaction took roughly 3,100 GPU-hours of computing, or about $320. This week the estimated cost is around $67. The technology didn’t change. The price fell because StarkWare turned the problem into a contest, and the internet — human and otherwise — showed up.
To understand why a Bitcoin transaction costs GPU-hours at all, you need the mechanism. The expensive part isn’t the Bitcoin fee; it’s a brute-force search that happens on the user’s own hardware before anything reaches the blockchain. The construction slots a hash into the place where Bitcoin normally expects a digital signature, and only about one in 70 trillion hash outputs has the required shape. So the software hashes inputs over and over again — trillions of times — until one of them fits. It is, in effect, an enormous lottery whose prize is permission to move your own money. Faster code means fewer GPU-hours to buy the same winning ticket.
StarkWare ran the Quantum-Safe Bitcoin Optimization Challenge alongside Yukon Research and Eigen Labs, opening the code to outside solvers to see how far down the number could go. Pretty far, it turns out. One core benchmark jumped from 146 million verified candidates per second to more than 820 million on a standard RTX 4090 — a nearly sixfold speedup in a week — with 62 improvements promoted across two tracks. Which is what lottery tickets cost when thousands of people are competing to print them faster.
The detail that will not surprise anyone who has watched software development lately is who won. The leading records were held by developers running AI models, with StarkWare citing Anthropic’s Opus 5 and Fable 5.1 at the top, and OpenAI’s GPT-6 Astra, Grok 4.6 and Kimi close behind. The competitive frontier of low-level cryptographic optimization is now, at least partly, a benchmark for large language models writing hash-guessing code. The models are presumably not worried about quantum computers; they are just very good at the contest.
StarkWare was careful to walk back any triumphalism. The $67 is an estimate under stated hardware assumptions, not a market price, and it reprices every time someone beats the record. The effort also doesn’t make Bitcoin quantum-safe on its own: these transactions are nonstandard and have to be delivered directly to a miner rather than broadcast normally, and the construction only shields coins whose public key hasn’t already been exposed. StarkWare itself maintains that a soft fork — an actual network-wide upgrade — remains the better long-term fix. This is a very clever bolt-on, and the company selling it would like you to know that.
The backdrop is the growing industry fretting about “Q-Day,” the hypothetical moment when quantum computers become powerful enough to break the elliptic-curve cryptography that secures Bitcoin wallets. Coinbase is among the firms already drafting post-quantum custody playbooks. Nobody knows when Q-Day arrives, or whether it arrives on schedule, but the precautionary market in hashing your way to safety is apparently clearing at around $67 and falling.
One in 70 trillion hashes works. A week ago it cost $320 to find one; now it costs $67, and the fastest searchers in the world are chatbots. Whatever quantum computers eventually do to Bitcoin, the race to defend it is already being run by machines.