Polymarket is tearing down the matching engine behind its onchain venue and rebuilding it in Rust. This is not a quiet remodel. It is a from-scratch rewrite aimed at mid-to-late fourth quarter, with early benchmarks that promise a sharp jump in speed.
On Aug. 31, Josh Stevens, vice president of DeFi engineering, posted that the DeFi central limit order book was “full of tech debt.” He said the book grew fast, was hacked together, and became inherited code. His team had already scaled it 10x and made it 3x faster. The foundations, he wrote, were still “wrong in almost every way.”
Patch one fault and another opened. Stevens’s conclusion: “the codebase can’t be saved long-term.” So the company is replacing the matcher even while keeping the live book running.
That dual track is the story through year-end. Claims of higher speed are one thing. Shipping a new matcher without scrambling signed-order flow is the hard part. Like rebuilding a busy freeway without closing it to traffic.
Why the Current CLOB Hit a Wall
Polymarket’s docs describe a simple loop. Traders sign orders, send them to the CLOB (Central Limit Order Book), the operator can match them, and fills settle on Polygon. Depth and live updates already flow from clob.polymarket.com.
Stevens said the loop is not the problem. The assembly is. Fixes in one layer kept creating failures in another. The book still works. But every new change costs more than the last.
He also made the business case. “The exchange is the heart of Polymarket; everything runs through it.” If the venue is going to hold peak load, he argued, the engine has to be fast, stable, and scalable. Patching a foundation never built for that traffic would keep producing the same latency spikes. A problem all engineers face when a system designed for lower traffic is suddenly asked to hold up under much greater volume.
Meanwhile, the old CLOB stays in maintenance. Bots and market makers already live inside tight posting and cancel caps, spelled out in Polymarket’s CLOB API rate-limit tables. A faster matcher does not automatically raise those caps. It only changes what happens after an order clears the gate.
The Rust Targets: Latency, Throughput, and No GC Stalls
Stevens published the numbers that stuck. Early tests show 10x to 20x faster p99 latency than today’s engine. Throughput is aimed at 200,000 orders per second, about 15x current capacity, with room toward 400,000-plus.
He tied the gains to Rust. The programming language offers memory safety without a garbage collector, so there are no random pauses and fewer surprise tail spikes. In his telling, the benchmark tail should match the production tail, because the compiler forces correctness before the code runs. Data races and memory corruption, he wrote, “can’t exist in the first place” unless engineers step into unsafe blocks.
That is the same pitch low-latency shops use when they pull hot paths out of garbage-collected languages. Independent Rust clients already talk to Polymarket’s current book with lock-free state and lean hot paths. Those tools do not replace the matcher. They only show how much of the surrounding stack moved first.
Lab multiples can still die on real-world contact with networking, signature checks, or settlement. Treat 200,000 orders per second as a design target until the new engine faces a huge market spike or a crowded short-dated book. This is another PolyPunter official “we shall see”.
The Rewrite Hits During a Broader DeFi Reset
In mid-September, CEO Shayne Coplan said onchain performance had “waned” as the company scaled and hired former Zora chief Jacob Horne to work with him on product, “in particular making Polymarket DeFi great again.”
Coplan also promised a DeFi town hall. He admitted some longtime users think attention shifted to the regulated U.S. venue; a not completely unjustified belief. The Rust rewrite is the most concrete engineering reply to that complaint, even though Horne’s job is product, not code.
Watch three things into Q4: whether the dual-run period stays clean, whether live p99 actually compresses toward the lab projected multiples, and whether post, cancel, and depth APIs stay stable enough that bots do not need an all-hands panic rewrite.
References
- Josh Stevens, X post on the Polymarket DeFi CLOB rewrite, Aug. 31, 2026
- The Defiant, “Polymarket Hires Ex-Zora CEO for Onchain Product Push”
- Polymarket documentation, Orderbook
- Polymarket documentation, API rate limits
- BeInCrypto, coverage of Jacob Horne joining Polymarket
- The Coin Republic, Horne hire
- Phemex News, Horne hire report
- Coinotag, Horne hire, and Polygon DeFi rebuild
- YouTube, “Rust for mission-critical software”
- Databento, “Rust vs C++ for trading systems”
