AI Agents Are Racing to Make Quantum-Safe Bitcoin Cheap—And Winning
An open contest run by StarkWare, Yukon Research, and Eigen Labs cut the estimated GPU cost to produce a quantum-safe Bitcoin transaction from about $320 to roughly $67 in one week by optimizing the client-side brute-force search. The fastest entrants used AI-assisted development, with Anthropic's Opus 5 and Fable 5.1 topping the leaderboards and other models like OpenAI's GPT-6 Astra, Grok 4.6, and Kimi also contributing significant improvements.
Why It Matters
The competition shows how software and machine-learning optimization can rapidly reduce the computational expense of a stopgap quantum-resistant transaction method, highlighting both the short-term practicality of such constructions and the broader urgency around Bitcoin's vulnerability to future quantum attacks.
Key Facts
- Organizers: StarkWare, Yukon Research, and Eigen Labs
- Estimated cost reduction: From about $320 to roughly $67 in GPU compute
- Initial mainnet transaction compute: ~3,100 GPU-hours, costing around $320
- Brute-force difficulty: Only about 1 in 70 trillion hash outputs fits the required shape
- Performance improvement: Benchmark rose from 146 million to over 820 million verified candidates/sec on an RTX 4090
An open optimization contest run by StarkWare together with Yukon Research and Eigen Labs drove a steep fall in the estimated GPU compute needed to craft a quantum-resistant Bitcoin transaction. Contest dashboard figures show an estimated cost drop from roughly $320 to about $67 over the course of a week, achieved by making the client-side brute-force hashing routine far faster.
The quantum-safe construction was first used on mainnet last month in a transaction developed by StarkWare; that initial build required around 3,100 GPU-hours and was estimated to cost about $320. The expensive phase of creating these transactions is a pre-blockchain brute-force search performed on the user's GPU: the scheme places a hash where Bitcoin normally expects a signature, and because suitable hashes are extremely rare—about one in 70 trillion—clients must hash inputs repeatedly until they find a fitting output.
Contest entrants focused on accelerating that hashing/search stage. StarkWare reported that a core benchmark increased from 146 million verified candidate hashes per second to more than 820 million per second on a standard RTX 4090, with 62 improvements promoted across two competition tracks. Leading submissions were produced with assistance from AI models; StarkWare named Anthropic's Opus 5 and Fable 5.1 as top performers, followed by OpenAI's GPT-6 Astra, Grok 4.6, and Kimi.
StarkWare cautioned that the roughly $67 figure is an estimate based on stated hardware assumptions and that it changes as better solvers appear. The company also emphasized these optimizations do not make Bitcoin inherently quantum-safe: the transactions created are nonstandard and must be relayed directly to miners, and the construction only protects coins whose public keys have not already been revealed. StarkWare continues to view a soft fork as a preferable long-term solution to quantum risk. The work arrives amid growing industry attention to the threat of quantum computers to elliptic-curve cryptography and related custody planning by firms such as Coinbase.
Keep Reading
OpenAI agent made unauthorized attempts to access federal agencies’ websites
Google Just Made Free 1080p AI Video Generation Available to Anyone
Control Resonant is a great game — it’s even better when you read everything
Meta turned Muse into a viral hit. Now comes the hard part.
Original source: Decrypt