Sui Live AI Agent Test Reaches 40.6M TPS With Tunneling

Sui has published results from a live stress test showing its programmable offchain “tunnels” can process transaction traffic at extremely high speeds, peaking at 40,614,180 transactions per second (TPS). The demonstration, held Wednesday at Sui Basecamp in Singapore, is intended to model the kind of high-frequency activity expected when large numbers of software agents interact continuously.

According to Sui, the test surpassed a prior benchmark recorded in July, when a comparable setup reached roughly 6.1 million TPS. In the new run, Sui said activity moved offchain through its tunnels and was settled to the Sui blockchain when tunnels were closed—allowing throughput to scale without forcing every individual transaction to be executed directly onchain.

Key takeaways

• Sui reports a live-tunnel throughput peak of 40,614,180 TPS, up from about 6.1 million TPS in a July test.
• The throughput came from offchain tunnel processing, with settlement to the blockchain occurring only when tunnels close.
• Sui opened 10,000+ tunnels on mainnet during the demo, with traffic including payments, games, and chat applications.
• CertiK is acting as an independent auditor and is reviewing the test data ahead of a forthcoming report.
• The company frames the work as relevant to AI agent-to-agent interactions rather than human-scale transaction volumes.

From July benchmark to 40.6M TPS peak

Sui’s latest numbers build directly on its earlier July throughput demonstration. In the October test at Sui Basecamp, the network said the offchain tunnels processed more than 40 million TPS during the stress window, with a reported peak at 40,614,180 TPS.

The headline improvement matters for a practical reason: it suggests Sui’s architecture can handle substantially higher traffic while using the same general tunnel concept—programmable channels that process activity offchain and then settle to the blockchain when closed. That settlement model is designed to reduce the burden of executing every operation onchain, which can otherwise cap throughput under heavy load.

How programmable tunnels change the execution model

The core of the demonstration is Sui’s programmable offchain tunnels—channels that move transactions offchain while maintaining an onchain settlement step at the end of a session. Sui described them as a mechanism to support high-frequency activity without requiring each transaction to be fully processed onchain in real time.

During the test, Sui said more than 10,000 tunnels were opened on Sui mainnet. The transaction stream spanned multiple categories, including payments, games, and chat applications. That breadth is notable because it implies the test was not limited to one narrow workload; instead, Sui attempted to reflect patterns common in interactive applications where message-like interactions can be frequent.

For developers and operators, the key question is not only whether the system can reach a number in a controlled stress test, but whether the tunnel model can support real application behavior under load. The separation between offchain processing and onchain settlement is intended to make that feasible—especially when the number of interacting entities grows beyond what human users typically generate.

Built for AI agent traffic, not human browsing

Sui’s framing of the results centers on a shift in how people expect digital systems to behave. Adeniyi Abiodun, co-founder and chief product officer at Mysten Labs and an original contributor to Sui, argued that traditional transaction targets do not align with agent-driven usage patterns.

Sui is built for the moment when millions of AI agents are transacting every day.

No market in the world needs 6 million transactions per second, let alone 40 million: people just don’t move that fast. Agents do.

The emphasis is a meaningful editorial point: high TPS claims can sound abstract if they are aimed at replacing user behavior rather than enabling new classes of machine-to-machine interactions. By tying the tunnel approach to AI agents, Sui is effectively positioning its throughput improvements as infrastructure for a potentially different traffic profile—one where message frequency and autonomous behavior could push systems into ranges that traditional benchmarks don’t cover.

Verification steps and what to watch next

Sui said CertiK served as an independent auditor during the demonstration and is reviewing the test data. The company indicated that an auditor report is expected in the coming days, which should help determine how the test was constructed, what was measured, and how results should be interpreted beyond the peak TPS figure.

That verification process is important for readers because stress tests can be influenced by implementation choices—such as workload design, tunnel lifecycle rules, and the exact criteria for what counts as a “transaction” within the tunnel pipeline. An independent review may clarify whether the 40.6M TPS peak reflects end-to-end throughput from application intent to onchain settlement, or whether it is primarily the offchain tunnel processing rate with later settlement.

As the audited report approaches, the market will likely focus on two practical questions: how closely the test workload maps to real AI-agent applications, and what performance trade-offs exist when tunnels close and settlement occurs. Until those details are reviewed, Sui’s results are best read as a strong signal of architectural capability under simulated high-frequency conditions rather than a guarantee of identical performance under all production scenarios.

In the meantime, investors, builders, and users should keep an eye on CertiK’s findings and on how Sui’s tunnel approach is described in subsequent technical documentation—especially around settlement behavior, real application workloads, and any constraints that might emerge outside controlled demos.