Polymarket Targets Mid-December Nex Launch After Faster Tests

Polymarket targets a mid-December launch for Nex, a Rust-based matching engine, Engineering Vice President Josh Stevens said. The company will run four more weeks of testing; internal tests reduced p99 order latency from 1.55 seconds to 3.76 milliseconds.

Nex has run against production traffic in shadow mode for one week. Polymarket will keep it there for four more weeks, with a mid-December launch planned “if testing goes well,” Stevens said.

The disclosed results cover order handling, cancellations and WebSocket writes. p99 cancel latency fell from 627.92 milliseconds to 3.20 milliseconds. p99 WebSocket write latency dropped from a prior range of 44.5 to 88.1 milliseconds to 0.099 milliseconds.

The p99 label marks the point at which 99% of observations fall at or below the stated latency, placing emphasis on slower responses rather than an average. These figures remain internal test results; Polymarket has not presented Nex performance from live production use.

Performance and operating capacity

Nex can process 100,000 instructions per second in steady state, compared with 10,000 per second for the existing system. It can reach a peak of 150,000 instructions per second. That puts sustained capacity at ten times the current steady-state rate and the peak rate at 15 times that benchmark.

Startup time fell from five minutes to 15 seconds, removing four minutes and 45 seconds from the engine’s startup window. The disclosure does not specify instance count, workload mix or sustained behavior at the 150,000-instruction peak, so the capacity figures still require validation outside the test environment.

Stevens said Polymarket will use the additional shadow period to assess Nex before a production release. The sequence gives Polymarket a defined observation window before deployment. A mid-December launch therefore remains possible rather than committed.

Polymarket has opened a Nex sandbox for user testing. The sandbox adds a user-facing test channel beyond Polymarket's internal measurements. Nex introduces no breaking changes to request and response formats, removing the need for a mandatory integration rewrite while preserving existing interfaces.

What does this change for market makers?

The immediate consequence is a shorter response window for automated order and cancel workflows, based on the reported latency figures. Market makers do not face a new request or response format under the plan, and the engine can absorb a larger sustained instruction flow.

These changes affect the execution infrastructure even though the disclosed results do not establish any shift in fill rates or market outcomes.

The WebSocket path may reach users first. Stevens said WebSocket improvements may roll out before the full engine, which would make part of the performance work available without waiting for the mid-December release. That staged deployment would not, by itself, deliver the full engine’s order-latency and throughput profile.

For practitioners, the remaining work is verification rather than interface migration. The disclosed results do not show how Nex behaves across a full trading cycle, under peak production load or once it becomes the live matching system. Four more weeks of shadow testing sit between those unanswered operating questions and deployment.

Polymarket’s next checkpoint is the end of four additional weeks of shadow-mode testing, with a mid-December launch possible if the results remain acceptable.