I sampled Base mainnet gas every 5 minutes for 7 days: 2,023 readings. 93% of them sat exactly on the fee floor of 0.006 gwei, and a single 25-minute spike on October 15 was enough to make 02:00 UTC the "most expensive hour" of the week.
The floor
Base gas is flat almost all the time. In 93.3% of readings the base fee sat at exactly 0.005 gwei and the gas price at 0.006 gwei; 98.3% were within 5% of that. Averaged over the week, gas ran just 1.3% above the floor.
The spikes
In 7 days gas left the floor by more than 5% in 23 separate events, most of them a single 5-minute reading. Only one was large: on October 15 the base fee jumped from 0.005 to 0.049 gwei and stayed elevated for about half an hour.
| Start (UTC) | Peak gas price | vs floor | Seen for | Uniswap swap at peak |
|---|---|---|---|---|
| Oct 15, 02:24 | 0.0499 gwei | 8.3x | ~25 min | $0.021 |
| Oct 13, 00:08 | 0.0137 gwei | 2.3x | ~10 min | $0.006 |
| Oct 13, 00:53 | 0.0099 gwei | 1.7x | ~5 min | $0.004 |
| Oct 14, 12:34 | 0.0091 gwei | 1.5x | one reading | $0.004 |
| Oct 17, 21:40 | 0.0086 gwei | 1.4x | one reading | $0.004 |
On the floor the same swap (180,000 gas) costs $0.0026 and a plain transfer $0.0003. Dollar figures use ETH at $2,387, the Chainlink price on October 17; durations are bounded by the 5-minute sampling, so a spike could have started or ended up to 5 minutes either side. The base fee drove every event: the median priority fee stayed near 0.001 gwei throughout.
Why a "cheapest hour" ranking misleads
Average the week by hour of day and 02:00 UTC comes out 21.6% above the floor, the most expensive hour by far. Leave out the hour of October 15 that held the spike (02:00 to 03:00 UTC) and the same hour drops to +0.2%, no different from any other.
The ranking also moves week to week. For September 23 to 30 the same method named 08:00 and 14:00 UTC as the hours to avoid; this week it named 02:00. On a chain that sits on its floor, an hour-of-day average mostly records where the week's one or two spikes happened to land.
What an agent should do with this
- Transact when the work is ready. On a normal hour the floor is the price. Waiting for a cheaper hour saved about 1.3% against the average hour this week, a fraction of a cent on a swap.
- Watch for spikes, not schedules. The risk is a short burst like October 15, not a daily rush hour. If a batch can wait, check whether gas is above the floor right now and retry in 15 to 30 minutes; every large spike this week was gone within half an hour.
- Distrust a single week's hour ranking. One event can make any hour look expensive. A ranking needs several weeks behind it before it says anything about a time of day.
- Budget in dollars. At $0.0003 a transfer and $0.0026 a swap, gas is rarely the cost that matters. Price it in dollars before deciding it is worth optimizing at all.
How the data was collected
The readings come from base-gas-x402, an open-source gas API I run on Base. Every 5 minutes it reads the latest block, eth_feeHistory over 10 blocks and eth_gasPrice from Base mainnet, and keeps 7 days of samples. The window here is October 11, 12:33 UTC to October 18, 12:32 UTC: 2,023 readings. Each gas price is base fee plus the network's suggested priority fee, in gwei; dollar figures are execution cost only, since Base's L1 data fee was under 1% of the total for these transactions.
The same data is available to agents, paid per call in USDC over x402, no API key:
npx agentcash@latest fetch "https://base-gas-x402-production.up.railway.app/gas/history?hours=168"
/gas/cheapest-window turns it into the answer above: the hours to avoid, the cheapest hour, and the saving against the average hour in percent and dollars, and it says so when timing saves nothing. There is also an MCP server and an agent skill. Site: base-gas-x402.vercel.app.


Top comments (0)