Smarter, Not Harder: How Flow Calculates Transaction Fees

Every blockchain transaction, whether minting an NFT or swapping tokens, consumes network resources. To keep the network secure and discourage spam, these resources must be priced. This is the purpose of transaction fees. While all blockchains have them, the method used to calculate fees has a profound impact on user cost and network performance.

With the Forte upgrade released in October 2025, Flow refined its unique approach to transaction metering. The update strengthened network resilience and improved fairness by aligning fees more closely with the resources each transaction consumes.

Flow’s Philosophy: Measuring What Matters

Flow uses a lightweight and highly effective heuristic to estimate the resource cost of transactions. This cost is divided into two components:

• Inclusion Fee reflects the resources required to admit a transaction into a block. It depends only on properties of the transaction text, such as its size or number of signatures, and can be calculated without executing it.
• Execution Fee covers the computational resources required to execute the transaction and verify its result.

Here’s where Flow’s approach stands out. Instead of meticulously counting every tiny operation, we recognize that most form a consistent and predictable “noise floor” of computation. The cost of these numerous lightweight operations can be estimated reliably from transaction size. Flow reserves detailed accounting for operations that truly affect performance, such as storage reads and writes, complex cryptographic functions, and loop iterations. This yields a cost model that is both accurate and efficient, without the overhead of counting everything.

A Tale of Two Models: Flow vs. Ethereum

To see the benefit of Flow’s approach, it helps to compare with Ethereum. The Ethereum Virtual Machine (EVM) charges gas for every instruction at a very fine-grained level. For example, computing c := a + b involves pushing a and b onto the stack (PUSH) and then adding them (ADD). The PUSH and ADD operations cost 3 gas units each. If you were computing c := a * b, the MUL operation would cost 5 gas.

However, these tiny instruction costs are insignificant compared to the expense of accessing blockchain state in EVM. Reading the number a from storage for the first time in a transaction (SLOAD) costs 2,100 gas. Storing the result c in a new storage slot (SSTORE) costs a massive 22,100 gas.

In our example, arithmetic operations contribute less than 0.02% to the overall cost. In practice, the values a transaction operates on are either loaded from the blockchain state or contained in the transaction, whose size is already accounted for in the inclusion fee. Research shows that simply tracking loop iterations plus a small set of expensive system function calls during execution yields a surprisingly precise estimate of the computational resources consumed.

Flow’s metering framework embraces this reality and focuses accounting on the operations that truly matter. This approach minimizes overhead, makes the network more sustainable, and keeps costs low for users. Just as important, it strengthens resilience: periodic recalibration keeps fees aligned with real-world resource consumption as hardware and software evolve, reducing the risk that outdated pricing could be exploited by malicious clients for resource exhaustion attacks. In fact, recalibration is a requirement for any blockchain system to maintain accurate pricing as execution environments evolve. Flow’s approach turns this necessity into a built-in process rather than an afterthought.

Our Latest Calibration - improving network resilience and fairness

A good model is a living model. As hardware advances and Flow’s Virtual Machine (FVM) and Cadence smart contract language continue to evolve, the real-world resource consumption of transactions naturally drifts away from any static model. Different operations benefit to varying degrees from optimizations, and new features change how resources are utilized. Periodic recalibration keeps fees fair for users and effective in protecting the network.

Execution Effort Calibration 2 is the most comprehensive analysis to date. Its goal was to determine how much the real-world resource consumption of FVM operations had diverged from the previous model and to calibrate a more accurate one. The earlier model, published in FLIP 753: Variable Transaction Fees – Execution Effort I, was intentionally kept compact. At the time, a simpler design was considered more robust against overfitting the rapidly evolving system characteristics. Since then, Flow’s runtime has matured significantly. For example, the 2024 Crescendo upgrade introduced Cadence 1.0, the first long-term stable Cadence release featuring substantial optimizations and many new operations. Both the Crescendo upgrade and the 2025 Forte upgrade also delivered major performance improvements across the protocol stack.

To quantify how far real-world resource consumption had drifted from the existing metering model, thousands of transactions were analyzed. The following scatter plot illustrates how accurately the previous model predicted execution time for synthetic transactions covering every operation type in the FVM, including operations not yet available when the model was created.

Over time, real-world execution times drifted away from the predictions of the old execution-effort model, yielding an R² of –47.7. The coefficient of determination R² measured how well the model’s estimates match observed data, where 1.0 represents a perfect fit. A negative value indicates that the system’s resource usage has evolved beyond what the old model can reliably capture. This result was expected: the model was calibrated more than two years earlier and was now tested against a significantly optimized and expanded system. If left uncorrected, such drift between modelled and actual resource use could open vectors for resource exhaustion attacks, underscoring the importance of regular recalibration to maintain network resilience.

In comparison, the following scatter plot illustrates the significantly improved accuracy of the recalibrated model, evaluated on the same set of transactions. It shows a much tighter correlation between predicted execution effort and real-world computation time than the old model.

The new model exhibits a clear, near-linear relationship between priced effort and observed execution time, with an R² of 0.73. Outliers are rare and less pronounced, confirming that the calibration successfully restored alignment between pricing and resource consumption. This reinforces both network resilience and fairness for all participants.

Looking ahead, we plan to advance this methodology further by:

• Enabling seamless recalibration whenever software changes affect execution behavior
• Automatically detecting outliers that could reveal model drift before it impacts the network
• Ultimately integrating model updates directly into the protocol itself, enabling dynamic adjustments rather than compiling model parameters into the node software as is done today. Flow’s dynamic protocol state provides a natural foundation for evolving the execution-effort model in this way.

What Flow’s Metering Approach Means for You

For developers and users on Flow, the calibration improvements in Forte bring real, measurable benefits:

• Lower Costs, Higher Throughput: By minimizing the computational effort spent on accounting, Flow unlocks more capacity for executing transactions. The result is exceptionally low fees for users and an unparalleled amount of computation that developers can pack into a single transaction.
• Improved Network Resilience: Regular calibration keeps fees aligned with real-world resource consumption, ensuring that transactions are priced fairly as hardware and software evolve. This alignment hardens the network against both intentional and accidental resource exhaustion attacks, making Flow more robust for everyone.
• Faster Innovation: Lightweight metering means less time spent maintaining complexity and more time building what matters. The Flow team can focus on pushing the boundaries of scalability, resilience, and developer experience across the network.

This calibration effort is a key part of the Forte upgrade and a perfect example of Flow's commitment to pragmatic, impactful engineering. Flow is a decentralized, fault-tolerant computing platform that combines scalability, efficiency, sustainability, and fairness to meet the most demanding real-world use cases.

‍