Every pricing decision starts with a question the agent has to answer before it touches a number: which machines am I actually competing against? Strategy modes are how you answer it per machine. The pricing logic stays the same, the floor stays enforced, but the comparison set changes, and with it how hard and against whom your listing competes.
Why the comparison set is the strategy
On Vast.ai, an RTX 4090 in a datacenter with fast storage, an RTX 4090 in a garage on residential internet, and four RTX 4090s sold as one instance all show up under the same GPU name. Price them against each other naively and someone gets hurt: the strong machine gets dragged down to garage prices, or the modest one prices itself out of its real market. Picking a strategy mode is picking the market your machine belongs to. The undercutting mechanics, described in bid offset explained, then operate inside that market.
The four modes
Market Match: the default
Market Match prices against comparable available and recently rented GPUs of your type, while respecting your floor and platform safeguards. It uses rented offers as the primary signal, targets the rented P50 median, undercuts the cheapest comparable competitor by your bid offset, and raises your price back toward the next competitor when you are already the cheapest.
This is the right mode for most machines most of the time: common consumer cards in ordinary configurations, competing in the broad pool where most renters shop.
Performance Match: compete on measured speed
Performance Match prices against Vast hosts whose cards actually perform like yours, rather than against every listing that shares the GPU name. Two machines with the same card on the label can benchmark far apart: power limits, cooling, PCIe topology, and CPU pairing all move real throughput.
If your machine benchmarks near the top of its GPU class, this mode stops it from being priced against the slowest listings wearing the same name. If it benchmarks lower, the mode is honest in the other direction and positions you with your true peers, which tends to win more rentals than pretending.
Premium Multi-GPU: price the whole rig
Premium Multi-GPU targets renters taking at least your selected GPU count and prices against that larger-rental market. Renters who need 4x or 8x machines are a different market from single-card renters: supply is thinner and clearing prices per GPU often run higher, because a machine that can serve them is rarer.
You choose the minimum GPU count to target, and you choose the pricing basis underneath: Market Match or Performance Match. The dashboard shows the combination, for example Premium 4x + Market Match. A four-card rig priced per card like four separate single-card rigs is usually underpriced for the renters it really serves; this mode is the correction.
Autosort Target: eyes on the ranking
Autosort Target is different in kind: it only observes. It watches where your machine lands in Vast.ai search results and what it would take to move up, while Market Match keeps setting your price. Use it when you want ranking intelligence about a machine, on top of normal pricing behavior, without handing position-chasing the keys.
Which machines suit which mode
| Machine | Sensible mode |
|---|---|
| Single consumer card, ordinary specs | Market Match |
| Well-tuned rig that benchmarks above its GPU class average | Performance Match |
| 4x or 8x rig renters take as one instance | Premium Multi-GPU, with the basis your single-card results would suggest |
| Any machine whose search ranking you want to understand | Autosort Target, alongside its normal Market Match pricing |
Modes are per machine, not per fleet
Read the results before changing the settings
Every decision the agent makes lands in the Activity feed with the market context that produced it: your price, the competitor it measured against, the median, and the clearing price. The action types are the vocabulary: Updated means the price moved and the entry shows why, Held means the current price was already correct for the market it saw, Skipped means a check ran but conditions ruled out a write, and Rented or Available track the machine's state changes. That context is how you evaluate a mode. If a Performance Match machine logs Held entries because its performance peers are all rented, that is the mode working. If a Premium Multi-GPU machine logs long idle stretches, the larger-rental market may be thin for your GPU type right now, and Market Match might fill more hours.
A useful weekly habit: filter the feed to one machine, scan the Updated entries, and ask whether each move is one you would have made yourself given the market context shown. When the answer is consistently yes, the mode fits the machine. When you keep disagreeing, the disagreement usually points at the comparison set, which is precisely what a mode change adjusts.
Change one machine's mode at a time and give it days, not hours, before judging: rental demand is lumpy. The activity log guide covers how to read each action type. If you are not set up yet, the quick start gets a machine from API key to automated pricing in about 3 minutes, and the Vast.ai hosting portal is where your listings live on the platform side.