Solana DEX failure rates, measured
Everyone quotes Solana fees. Almost nobody publishes what share of swaps actually fail, and the answer differs by an order of magnitude between venues. These figures are read straight out of finalized blocks, not estimated, and they are re-measured every week.
The short version. Across 776 transactions sampled from 1 finalized blocks over roughly 16 minutes, 3.2% of all Solana transactions failed. By venue the spread is far wider: Pump.fun curve failed 45.5% of the transactions that touched it, while PumpSwap failed 38.8%. That difference is the single most under-discussed cost in this category, because a failed swap still pays its fee and produces no volume.
Failure rate by venue
| Venue | Transactions | Failed | Failure rate | Median fee |
|---|---|---|---|---|
| PumpSwap | 49 | 19 | 38.8% | 0.000005 SOL |
| Meteora DLMM | 13 | 4 | 30.8% | 0.000008 SOL |
| Pump.fun curve | 11 | 5 | 45.5% | 0.000005 SOL |
A transaction is counted for a venue when that venue's program appears in the transaction's account list. Aggregator routes therefore appear against the underlying pool as well as the aggregator, which is the correct way to read them for execution purposes.
What people actually pay in priority fees
Priority fee guidance is usually a recommendation. This is the distribution of what was really paid, derived from the fee field of each transaction minus the fixed 5,000 lamport base fee.
| Percentile | Priority fee paid | In SOL |
|---|---|---|
| Median (p50) | 0 lamports | 0 SOL |
| p75 | 0 lamports | 0 SOL |
| p90 | 0 lamports | 0 SOL |
| p99 | 5,000 lamports | 0.000005 SOL |
91.1% of sampled transactions paid no priority fee at all. That is the number worth sitting with: the fee market is not a flat toll, it is a long tail, and the difference between the median and the p99 is several orders of magnitude. Sizing a campaign against the p99 wastes money; sizing against the median gets you outbid during the exact minutes that matter. The fee breakdown guide works through where this sits in a full campaign budget.
The live fee market
For comparison, the estimator's recommended levels for a busy launchpad program at the time of measurement, in micro-lamports per compute unit:
| Level | Micro-lamports per CU |
|---|---|
| Low | 0 |
| Medium | 510 |
| High | 56,025 |
| Very high | 1,684,868 |
Network conditions during the sample
- Median throughput: 4,296 transactions per second, p90 4,618.
- Median slot time: 267 ms.
- Median transaction fee across all traffic: 0.000005 SOL, p90 0.000005 SOL.
Week over week
| Measured | Overall failure rate | Median fee | Median TPS | Priority p90 |
|---|---|---|---|---|
| 2026-09-27 | 3.2% | 0.000005 SOL | 4,296 | 0 lamports |
| 2026-09-21 | 9.2% | 0.000005 SOL | 4,952 | 1,133 lamports |
| 2026-09-14 | 13.8% | 0.000005 SOL | 3,965 | 3,207 lamports |
| 2026-09-08 | 17% | 0.000005 SOL | 3,716 | 3,000 lamports |
| 2026-09-02 | 28.3% | 0.000005 SOL | 3,896 | 5,000 lamports |
| 2026-08-15 | 13.3% | 0.000005 SOL | 2,792 | 1,314 lamports |
How this is measured
A scheduled job asks a Solana RPC node for the current finalized slot, then pulls a set of blocks spaced several hundred slots apart so the sample covers a window rather than a single moment. For every transaction in those blocks it records whether the metadata carries an execution error, what fee was charged, and which known DEX programs appear in the account list. Percentiles are computed over the raw values, not over bucketed averages.
What this does not claim: it is a sample, not a census, and a sample taken during unusual conditions will read differently from one taken on a quiet afternoon. Venue attribution counts transactions that touch a program rather than trying to decode instruction data, so a transaction that routes through two venues is counted for both. Figures are republished on every measurement rather than smoothed, which means an unusual week will look unusual here. Our measurement policy covers what else we refuse to estimate.
The raw snapshot behind this page is available as JSON, including the block range and sample sizes, so any figure here can be checked or recomputed.
Every measurement is also archived in a public repository as JSON and CSV, under an MIT licence, alongside the program ids used for venue attribution: Solana DEX Reliability Index on GitHub. Anyone is free to reuse the figures; please quote the measurement date with them, because they move.
A written report covering the current sample, the method, a comparison of two measurements taken under different network load, and the full list of limitations is available as a PDF: Solana DEX Reliability Index, August 2026 (4 pages).
Why this matters for a campaign
A failure rate is not an abstraction. If a venue rejects a large share of the transactions sent to it, then a campaign routed there pays base and priority fees on every attempt and books volume only on the ones that land. Two campaigns with identical budgets can produce very different volume figures purely because of where they were routed, and that is a decision made before anything is submitted, not after. That is exactly why the console detects the venues holding a pool for your token first and routes across them rather than concentrating on one, which the troubleshooting guide covers from the other direction.
Route around the venues that reject you
Paste a mint address and the console shows every pool holding your token, the estimated swap count and the exact SOL fee before anything runs.
Open the volume console