Sui Live AI agent test reaches 40.6 million TPS in tunnel operation

Sui Programmable off-chain tunnels set new throughput records in stress tests to simulate real-time communications among AI agents

Sui announced that its programmable off-chain tunnels reached a new throughput milestone in an on-site stress test designed to simulate high-intensity, real-time communications between artificial intelligence (AI) agents. Network data shows that peak transaction processing speeds reached 40,614,180 transactions per second (TPS), much higher than the approximately 6.1 million TPS achieved in similar tests in July this year.

The demonstration was held at the SuiBasecamp event in Singapore on Wednesday. According to Sui, the activity is performed offline through "tunnels" that can batch process or manage high-frequency operations and settle the results to the Sui blockchain when the channel is closed.

Core Points

• Peak performance: Sui reported that field test peaks for tunnels under the programmable chain reached 40.6 million TPS, significantly higher than the July benchmark of approximately 6.1 million TPS.
• Scale and scope: During the event, more than 10,000 tunnels were launched on the Sui main network, with traffic covering multiple application categories.
• Design Purpose: Tunnel is designed to support high-frequency inter-agent interactions without the need to directly link every microtransaction.
• Independent Audit: CertiK participated in this test as an independent auditor and is reviewing the data and is expected to release an audit report soon.

Test principle: off-chain processing and on-chain settlement

Sui's core proposition lies in its "programmable off-chain tunnel", a mechanism designed to achieve fast transaction processing outside the main chain while retaining necessary on-chain settlement steps. In practical applications, this tunnel model is seen as a means to keep computing and transaction loads controllable despite extremely high frequency activity.

During the stress test, Sui stated that transactions were processed through these tunnels and then settled the results to the Sui blockchain when the channel closed. This design is critical because it targets a common bottleneck in high-throughput systems: Even if the network is able to perform a large number of transactions, if every operation is required to be performed directly on the chain, this often becomes a limiting factor when throughput requirements surge.

Scale and scope of on-site stress testing

Sui reported that the test opened more than 10,000 tunnels on its main network. Networks also said transactions cover scenarios such as payments, games and chat apps-use cases that often benefit from fast, interactive message flow.

According to Sui, the primary indicator of the test was a peak throughput of 40,614,180 TPS. For comparison, Sui pointed out that under similar test settings, attempts in July this year reached approximately 6.1 million TPS.

This growth is not only an engineering milestone, but also demonstrates that Sui's off-chain tunnel approach can scale at higher transaction rates than earlier benchmarks. However, investors and developers may still need more details to clarify how the tunnel will perform under different transaction portfolios, tunnel life cycles, and settlement models-these details are usually disclosed only after analysis of test data is completed.

Why Sui linked the results to AI agent activities

Sui's marketing perspective is that the tunnel mechanism is built for future scenarios where "millions of AI agents" may need to continue trading. Adeniyi Abiodun, co-founder and chief product officer of Mysten Labs (also described as an original contributor to Sui), believes that traditional networks are not optimized for such transaction volume.

Abiodun said in his speech that the world does not need the processing power of millions of TPS on human-driven activities, but agents do. His view essentially refers to the emergence of a mismatch between demand patterns and the throughput requirements that emerging agent applications may generate.

The practical implications for users and developers are that agent-focused applications may require infrastructure that can absorb bursts of rapid interaction traffic while keeping on-chain workload within a limited range. Sui's view is that off-chain tunnels with on-chain settlement functions are one way to achieve this balance.

Independent audit in progress: CertiK will review data

Sui said CertiK acted as an independent auditor in this demonstration and is currently reviewing test data. The network added that the report is expected to be released in the next few days.

This is important because high-throughput claims in the blockchain ecosystem often rely on test methodology-that is, exactly what is counted as a "transaction", which calculations occur offline rather than on the chain, and how settlements are processed. Audit reviews help reduce uncertainty and confirm whether the headline TPS reflects comparability with previous benchmarks or whether the test settings have changed significantly that affects results.

Although the comparison between Sui and July tests provides a reference point, readers should still pay attention to how CertiK describes the performance characteristics of the system and whether the test results hold true under conditions other than the demonstration environment.

Focus on follow-up observations

As CertiK's review approaches completion, the next question is how detailed audit findings will describe tunnel behavior and settlement costs-and whether the peak of 40.6 million TPS can be consistently reproduced outside of a controlled stress testing environment. For developers of agent-driven applications, the key is turning throughput milestones into predictable performance under real-world workloads.