Sui says its programmable offchain tunnels hit a new throughput milestone during a live stress test meant to mimic heavy, real-time communication between AI agents. The network reported a peak of 40,614,180 transactions per second (TPS), substantially above the roughly 6.1 million TPS achieved in a similar test in July.

The demonstration took place on Wednesday at Sui Basecamp in Singapore. According to Sui, the activity was carried out offchain via “tunnels” that can batch or manage high-frequency operations and then settle to the Sui blockchain when the channels are closed.

Key takeaways

Sui reported a live-test peak of 40.6M TPS for programmable offchain tunnels—well above its July benchmark of about 6.1M TPS.
More than 10,000 tunnels were opened on Sui mainnet during the event, with traffic spanning multiple application categories.
The tunnels are intended to support high-frequency agent-to-agent activity without forcing every micro-transaction to be executed directly onchain.
CertiK participated as an independent auditor and is reviewing the test data, with a report expected soon.

What Sui tested: offchain tunnels that settle onchain

Sui’s claim centers on its “programmable offchain tunnels,” which are designed to enable rapid transaction processing outside the main chain, while still preserving an onchain settlement step. In practice, the network frames the tunnel model as a way to keep computational and transaction load manageable when activity is extremely frequent.

During the stress test, Sui said it processed transactions through these tunnels and then settled the results to the Sui blockchain once the channels were closed. That design matters because it targets a common bottleneck in high-throughput systems: even if a network can execute many transactions, the requirement to handle every operation directly onchain can become the limiting factor when throughput needs surge.

Scale and scope of the live stress test

Sui reported that the test involved the opening of more than 10,000 tunnels on its mainnet. The network also said the transactions encompassed scenarios such as payments, games, and chat applications—use cases that typically benefit from fast, interactive message flows.

The test’s headline metric, according to Sui, was a peak throughput of 40,614,180 TPS. For context, Sui compared this outcome to an earlier attempt in July that reached approximately 6.1 million TPS under a similar testing setup.

That jump is significant not just as an engineering milestone, but because it suggests Sui’s offchain tunnel approach is scaling across a much larger transaction rate than in the earlier benchmark. Still, investors and builders will likely want more clarity on how the tunnels behave under different transaction mixes, tunnel lifetimes, and settlement patterns—details that are often only revealed after test data is analyzed.

Why Sui connects the result to AI agent activity

Sui’s marketing angle is that the tunnel mechanism is built for a future where “millions of AI agents” may need to transact continuously. Adeniyi Abiodun, co-founder and chief product officer at Mysten Labs and described as the original contributor to Sui, argued that conventional networks are not optimized for this kind of volume.

In his remarks, Abiodun said the world does not require multi-million TPS for human-driven activity, adding that agents do. His point is essentially about the mismatch between current demand patterns and the throughput envelope that emerging agentic applications could generate.

For users and developers, the practical implication is that agent-heavy applications may need infrastructure that can absorb bursts of rapid interactions while keeping onchain work constrained. Sui’s message is that offchain tunnels with onchain settlement are one path to that balance.

Independent audit underway: CertiK to review the data

Sui said CertiK served as an independent auditor during the demonstration and is reviewing the test data. The network added that a report is expected in the coming days.

That matters because high throughput claims in blockchain ecosystems often depend on test methodology—what was actually counted as a “transaction,” what computation occurred offchain versus onchain, and how settlement was handled. An audit review can help reduce uncertainty about whether the headline TPS reflects an apples-to-apples comparison with prior benchmarks or whether the setup changed in ways that materially affect the results.

While Sui’s comparison to the July test provides a reference point, readers should still watch for how CertiK characterizes the system’s performance characteristics and whether the test results hold under different conditions than the demonstration environment.

What to watch next

With CertiK’s review expected soon, the next question is how the detailed audit findings describe tunnel behavior and settlement overhead—and whether the 40.6M TPS peak can be consistently reproduced outside a controlled stress-test setting. For builders targeting agent-driven applications, the key will be translating throughput milestones into predictable performance under real workloads.