API Documentation

Conventions#

• Requests: GET on https://go.runs-on.com, except POST /api/facets; every GET endpoint answers HEAD too. Responses are JSON unless stated otherwise, and allow every origin (Access-Control-Allow-Origin: *).
• Errors: a JSON object {"error": "..."} whose message names the faulty parameter. 400 for an invalid parameter, 404 for an unknown endpoint, instance type or family, 405 for another method (with an Allow header), 500 when the database fails.
• Units: prices in USD per instance-hour, rounded to 6 decimals; memory in GiB; instance storage in decimal GB, as AWS sizes it; network bandwidth in Gbit/s; EBS bandwidth in Mbit/s; times in RFC 3339, UTC.
• Absent fields: a field that does not apply (no GPU, no instance storage, no PassMark score, no spot price) is left out of the object, not set to null, unless its table says otherwise. Test for its presence.
• Platforms: linux (or linux/unix), windows and macos (or mac), in any case. Responses name them Linux/UNIX, Windows and macOS.
• Zones: prefer zone IDs such as use1-az4. AWS maps zone names such as us-east-1b to physical zones separately in each account, so a name here may be another zone in yours. Your instance's zone ID is in its metadata (placement/availability-zone-id).
• Coverage: the instance types AWS offers in each region the site's AWS account can use. Prices refresh about every 90 minutes. The finder, facets and compare endpoints list the types priced in the last 7 days.
• Rate limits: none, but please be gentle and cache: an unfiltered /api/finder answer holds every priced type, a few megabytes.

Price figures#

The pricing fetcher records the current spot price of every availability zone (a zone below) and the on-demand price of every region; each of those passes is a run. Runs start 90 to 95 minutes apart, about 15 a day. Runs are told apart by gaps of more than 15 minutes between rows. The figures below cover the requested regions and platforms (Linux/UNIX unless you pass platform) over the 7 days that end at the start of the current 10-minute window (UTC, starting at :00, :10, :20...): prices recorded since count from the next window, and a run counts in the typical spot figures from the first window that starts at least 10 minutes after its last row. So the finder, facets and compare endpoints, the family exports and the site's pages move to a new run together.

Three kinds of spot price never count in the spot figures below, in the averages of /api/instances or on the site's pages:

• $0 prices: the type has no spot in that zone (macOS has none), or the spot fetch failed.
• Zones AWS stopped pricing: no AWS spot price for more than 3 days, measured from the newest price of their region and platform. AWS publishes the price of a live zone every few hours, and its history keeps returning the last price of a zone it stopped pricing.
• Zones AWS no longer gives to new accounts: today usw1-az2 in us-west-1, which sells old types (m1, m2, m4, r4, c1) at 10% of on-demand to the accounts that already had it.

Their on-demand prices and their regions still count. /api/instances/{type} lists them, flagged.

Spot figures

• typicalSpotPrice - What spot typically costs: for each run of the last 7 days, the spot price of the cheapest zone of the requested regions (and platforms), averaged over those runs, after the glitch rule below. Quote this figure: it moves little from day to day. A missing value means unknown, not "no spot": the type had no counted spot price in a settled run (a type without spot, macOS, a type first priced minutes ago), or the figure could not be computed. Fall back to avgSpotPrice, which is 0 when no spot price counts at all.
• typicalSpotRuns - Runs averaged in typicalSpotPrice: about 15 a day, so about 105 for a full week, fewer for a type that was not priced in every run.
• typicalSpotSince - Start of the oldest run averaged.
• typicalSpotRegion, typicalSpotZoneId - The zone that was the cheapest in the most runs, where typicalSpotPrice mostly comes from; typicalSpotZoneRuns is in how many of the typicalSpotRuns. A run whose cheapest zone the glitch rule skipped counts for no zone; a tie goes to the first zone in region and zone ID order.
• spotPriceNow - Spot price of the cheapest zone in the newest run, after the glitch rule.
• spotZoneAvgNow - Mean spot price across the zones of the newest run, without a zone skipped by the glitch rule.
• spotPriceNowAt - Start of the newest run.
• Glitch rule - A run's cheapest zone is skipped, in favour of the run's next cheapest zone, when it is below half of that next zone and below half of the cheapest price of each neighbouring run. The oldest and the newest run of the 7 days have one neighbour and are checked against it alone, so a real drop of one zone to below half of every other price shows in spotPriceNow from the run after it. A run with a single zone and a type with a single run are never checked.
• avgSpotPrice - Mean of every counted spot price of the 7 days: all zones of the requested regions, all runs. 0 when none counts.
• minSpotPrice, maxSpotPrice - Lowest and highest spot price of a single zone and run in the 7 days, with no glitch rule, and the region (minSpotRegion, maxSpotRegion) and platform (minSpotPlatform, maxSpotPlatform) it was seen in. minSpotPrice is a one-off low, below what spot typically costs.

On-demand figures

• avgOnDemandPrice - Mean of the requested regions' on-demand prices, each region counted once: AWS sets one on-demand price per region, so a region with six zones weighs no more than one with two. A region's price is its mean over the 7 days. 0 when AWS lists none.
• minOnDemandPrice, maxOnDemandPrice - The cheapest and dearest region's price, with that region (minOnDemandRegion, maxOnDemandRegion) and platform (minOnDemandPlatform, maxOnDemandPlatform).

Region counts

• regions - The requested regions in which the type has a spot or an on-demand price in the 7 days.
• spotRegionCount - How many of them have a spot price that counts: the regions behind avgSpotPrice and the other spot figures. A type can be sold in a region without spot.
• onDemandRegionCount - How many of them have an on-demand price: the regions avgOnDemandPrice averages.

Several platforms

With several platforms, or platform=all, each platform keeps its own figures in platformPrices, while the top-level figures mix them: avgSpotPrice and spotZoneAvgNow average the platforms' prices together, and avgOnDemandPrice each platform's regions (Windows prices include the license, so this inflates them); minSpotPrice, minOnDemandPrice, typicalSpotPrice and spotPriceNow take the cheapest price of any of them, in practice the Linux price; the region counts count a region once, whatever its platforms. Read platformPrices, or ask for one platform per request.

Find instances#

GET /api/finder

Every instance type priced in the last 7 days that matches the filters, with its specs and prices, and three Top 3 rankings. Every parameter is optional.

Parameters

A range is min-max (both bounds included), one value, or min+ for no maximum. In a raw query string a + decodes to a space, which reads the same, or send it as %2B. An empty value or all means no filter. An unknown value, a malformed, negative, NaN or reversed range, a decimal cpu or gpu value and a malformed region name are a 400; a well-formed region the site does not price matches nothing.

Examples

curl -sG "https://go.runs-on.com/api/finder" | jq '.results | length'

arm64 types with 4 to 8 vCPUs in us-east-1, the most PassMark per dollar first:

curl -sG "https://go.runs-on.com/api/finder" \
  -d region=us-east-1 -d cpu=4-8 -d arch=arm64 \
  | jq -c '.top3[] | {instanceType, passmark, typicalSpotPrice}'
{"instanceType":"c6gd.2xlarge","passmark":1105,"typicalSpotPrice":0.0317}
{"instanceType":"c8g.xlarge","passmark":1945,"typicalSpotPrice":0.0582}
{"instanceType":"m9gd.xlarge","passmark":2475,"typicalSpotPrice":0.0784}

64 vCPUs or more, in both us-east-1 and us-west-2:

curl -sG "https://go.runs-on.com/api/finder" \
  -d region=us-east-1,us-west-2 --data-urlencode "cpu=64+" | jq '.results | length'

The c8 and m8 families, the -flex families, one family's types:

curl -sG "https://go.runs-on.com/api/finder" -d region=us-east-1 -d "family=c8,m8" | jq '.results | length'
curl -sG "https://go.runs-on.com/api/finder" -d region=us-east-1 -d "family=*-flex" | jq '[.results[].instanceFamily] | unique'
curl -sG "https://go.runs-on.com/api/finder" -d region=us-east-1 -d "family=m7a.*" | jq '[.results[].instanceType]'

Types with AWS Inferentia or Trainium, and Windows prices:

curl -sG "https://go.runs-on.com/api/finder" -d accelerator=neuron | jq '[.results[].instanceType]'
curl -sG "https://go.runs-on.com/api/finder" -d region=us-east-1 -d platform=windows -d cpu=2 | jq '.top3ByPrice[0]'

The typical spot price of one type, and the zone it mostly comes from:

curl -sG "https://go.runs-on.com/api/finder" \
  -d region=us-east-1 -d platform=linux -d family=m8a.large \
  | jq '.results[0] | {instanceType, typicalSpotPrice, typicalSpotRuns, typicalSpotZoneId, spotPriceNow, spotZoneAvgNow}'

Response fields

A type without a price on that basis is left out of the three lists, and a type without a PassMark score out of the last two. Ties go to the cheaper type, then as in results. A list holds fewer than 3 types when fewer qualify, and top3PriceBasis is spot when nothing matches.

m8a.large in us-east-1, as of 2026-10-08, trimmed:

{
  "results": [
    {
      "instanceType": "m8a.large",
      "instanceFamily": "m8a",
      "provider": "ec2",
      "architecture": "x86_64",
      "vcpus": 2,
      "memoryGiB": 8,
      "gpus": 0,
      "cpuSpeedGhz": 4.5,
      "passmark": 3801,
      "cpuModel": "AMD EPYC 9R45",
      "cpuManufacturer": "AMD",
      "networkPerformanceGbps": 12.5,
      "networkBaselineGbps": 0.937,
      "networkBurstable": true,
      "networkPerformanceGiBps": 12.5,
      "ebsMaxBandwidthMbps": 10000,
      "ebsMaxIOPS": 40000,
      "ebsBaselineBandwidthMbps": 650,
      "ebsBaselineIOPS": 3600,
      "avgSpotPrice": 0.04266,
      "avgOnDemandPrice": 0.12172,
      "minSpotPrice": 0.0365,
      "minSpotRegion": "us-east-1",
      "minSpotPlatform": "Linux/UNIX",
      "maxSpotPrice": 0.0488,
      "maxSpotRegion": "us-east-1",
      "maxSpotPlatform": "Linux/UNIX",
      "minOnDemandPrice": 0.12172,
      "minOnDemandRegion": "us-east-1",
      "minOnDemandPlatform": "Linux/UNIX",
      "platformPrices": [...],
      "regions": ["us-east-1"],
      "spotRegionCount": 1,
      "onDemandRegionCount": 1,
      "passmarkSource": "measured",
      "passmarkSourceFamily": "m8a",
      "passmarkMeasuredAt": "2026-10-08T00:00:00Z",
      "passmarkBasis": "ec2-cpu",
      "typicalSpotPrice": 0.0365,
      "typicalSpotRuns": 1,
      "typicalSpotSince": "2026-10-08T20:44:17Z",
      "spotPriceNow": 0.0365,
      "spotZoneAvgNow": 0.04266,
      "spotPriceNowAt": "2026-10-08T20:44:17Z",
      "typicalSpotRegion": "us-east-1",
      "typicalSpotZoneId": "use1-az5",
      "typicalSpotZoneRuns": 1
    }
  ],
  "top3ByPrice": [...],
  "top3ByPerformance": [...],
  "top3": [...],
  "top3PriceBasis": "spot"
}

Instance type fields#

The fields of each result of /api/finder and /api/facets, of their Top 3 lists, and of /api/compare. /api/instances/{type} uses the same spec fields. "May be absent" fields are left out when they do not apply.

Type and CPU

Memory, GPUs and accelerators

Storage and network

Prices

Defined in Price figures, for the requested regions and platforms.

PassMark score#

passmark is the family's PassMark single-thread score, measured by RunsOn's open benchmark harness (runs-on.com/benchmarks). Every size of a family has the same score. The finder, the compare page, the family pages and the ranking show the same figure.

• Measured: runs-on.com publishes one figure per benchmarked instance type (passmarkSingle in ec2-cpu.json). A family's score is the median of its benchmarked types' newest figures (passmarkBasis ec2-cpu), or, without such figures, of its benchmark runs of the last 180 days (runs-median); with an even count, the higher of the two middle values (i7i's types score 3021, 3054, 3105 and 3239: i7i scores 3105). A measured score stays after its runs expire, with its date.
• Estimated: a family without benchmarks takes the score of a measured family with the same AWS processor: same manufacturer, same architecture and a sustained clock in common. The source is, in order, a fixed pairing (x2gd, g5g, i4g, im4gn and is4gen from m6g; i4i, x2idn and x2iedn from m6i; x2iezn from m5zn; x8aedz from m8azn), a sibling (r8gd from r8g), or a family of the same generation, preferring the m family (c9g from m9g). A few scores recorded before October 2026 are estimated with an unknown origin: they have no passmarkSourceFamily.
• None: a family whose processor no measured family shares has no score, and no passmark field.

The provenance fields are passmarkSource, passmarkSourceFamily, passmarkMeasuredAt and passmarkBasis (see Type and CPU). cpuSpeedGhz is AWS's stated clock, not a benchmark.

Faceted search#

POST /api/facets

The query behind the finder page: the results and Top 3 lists of /api/finder for a JSON body of filters, with the number of matching types per family and every value of each facet. Its facets follow the page and may change; for scripts, prefer /api/finder.

Request body

A JSON object of at most 64 KiB; every field is optional. Ranges take the forms of /api/finder, and several ranges of one field select the types in any of them.

An invalid JSON body or value is a 400, a larger body a 413, another method than POST a 405.

Response fields

Each facet item:

curl -s -X POST "https://go.runs-on.com/api/facets" \
  -H "Content-Type: application/json" \
  -d '{"regions":["us-east-1"],"architectures":["arm64"],"cpuBuckets":["4-8"],"text":"m8"}' \
  | jq '{familyCounts, top3: [.top3[].instanceType], cpus: .availability.cpus[2:4]}'
{
  "familyCounts": {"m8g": 2, "m8gb": 2, "m8gd": 2, "m8gn": 2},
  "top3": ["m8g.xlarge", "m8gd.xlarge", "m8gn.xlarge"],
  "cpus": [
    {"value": "4", "label": "4", "min": 4, "max": 4, "selected": true, "available": true},
    {"value": "8", "label": "8", "min": 8, "max": 8, "selected": true, "available": true}
  ]
}

Price histograms

A price histogram counts the types that match every filter but that price's own range: on the typical spot price (avgSpotPrice without one) for spotPriceHistogram, on avgOnDemandPrice (the cheapest platform's, with several) for onDemandPriceHistogram. The other price's range applies, so a range of one price narrows the other's histogram, never its own. Bins leave out their maximum, as the ranges do: the bins between two edges hold the types a range between them keeps.

A typical spot range off the ladder, in the m8 families: 0.033 is an edge, and the bins from 0.033 to under 0.07 hold the 15 results.

curl -s -X POST "https://go.runs-on.com/api/facets" \
  -H "Content-Type: application/json" \
  -d '{"text":"m8","spotPriceBuckets":["0.033-0.07"]}' \
  | jq -c '{results: (.results | length), spot: (.availability.spotPriceHistogram | {min, max, count, edges: .edges[8:15], counts: .counts[8:14]})}'
{"results":15,"spot":{"min":0.0048,"max":6.2929,"count":160,"edges":[0.025,0.03,0.033,0.04,0.05,0.06,0.07],"counts":[5,1,3,4,4,4]}}

A mistyped region, in OR mode (no counts):

curl -s -X POST "https://go.runs-on.com/api/facets" \
  -H "Content-Type: application/json" \
  -d '{"regions":["us-east-1","us-esat-1"],"regionMode":"OR"}' \
  | jq -c '.availability.regions[] | select(.selected)'
{"value":"us-east-1","label":"us-east-1","selected":true,"available":true}
{"value":"us-esat-1","label":"us-esat-1","selected":true,"available":true,"unpriced":true}

Compare instance types#

GET /api/compare

Two or three instance types side by side, as on the compare page, with the prices of /api/finder.

Parameters

Response fields

Each result adds:

Each missing item:

curl -sG "https://go.runs-on.com/api/compare" \
  -d instances=m7a.large,m7g.large -d region=eu-west-3 \
  | jq '{results: [.results[] | {instanceType, cpuModel, passmark, typicalSpotPrice}], missing, regions, suggestions}'
{
  "results": [
    {
      "instanceType": "m7g.large",
      "cpuModel": "AWS Graviton3 (Neoverse-V1)",
      "passmark": 1557,
      "typicalSpotPrice": 0.0417
    }
  ],
  "missing": [
    {
      "instanceType": "m7a.large",
      "instanceFamily": "m7a",
      "reason": "region",
      "regions": ["ap-northeast-1", "ap-southeast-2", "eu-central-1", "eu-north-1", "eu-west-1", "eu-west-2", "us-east-1", "us-east-2", "us-west-2"]
    }
  ],
  "regions": ["eu-west-3"],
  "suggestions": ["m7i.large", "m8g.large", "m6a.large", "m6g.large"]
}

Without a region the prices cover only the regions shared by every compared type, so they can differ from /api/finder's for the same type.

List instance types#

GET /api/instances

Every known instance type in name order, with its average prices over the last 24 hours per platform. These are plain means over every zone and run (for spot, of the prices that Price figures counts), not the typical spot price: for prices to quote, use /api/finder.

Parameters

An unknown platform or architecture is a 400, with /api/finder's message; an unknown region, zone or type matches nothing. Without a region, zone, platform or lifecycle=spot, the list includes the types without a price, with empty regions, platforms and pricing.

Result fields

curl -sG "https://go.runs-on.com/api/instances" -d type=m7a.xlarge
{
  "results": [
    {
      "instanceType": "m7a.xlarge",
      "instanceFamily": "m7a",
      "vcpus": 4,
      "memoryGiB": 16,
      "gpus": 0,
      "cpuSpeed": 2875,
      "passmark": 2875,
      "cpuSpeedGhz": 3.7,
      "cpuModel": "AMD EPYC 9R14",
      "cpuManufacturer": "AMD",
      "regions": ["ap-northeast-1", "ap-southeast-2", "eu-central-1", "eu-north-1", "eu-west-1", "eu-west-2", "us-east-1", "us-east-2", "us-west-2"],
      "platforms": ["Linux/UNIX", "Windows"],
      "pricing": {
        "Linux/UNIX": {"avgSpot": 0.103587, "avgOnDemand": 0.256712},
        "Windows": {"avgSpot": 0.221, "avgOnDemand": 0.440712}
      },
      "links": {"self": "/api/instances/m7a.xlarge"}
    }
  ]
}

Zone prices of an instance type#

GET /api/instances/{type}

The latest spot and on-demand price of each zone and platform for one instance type (m7a.xlarge, or m7a.xlarge.json), or for every type of a family (m7a.* or m7a.). Names are lower case: another spelling, and a trailing slash, redirect (301) to the lower-case path, with its .json and its query. A name without a dot is the family export.

Parameters

Each result is the latest price of the last 24 hours for one zone and platform. Every zone is listed, so that you can look up your own, in this order:

• the spot prices that count in the finder's figures, lowest first;
• then the prices that do not (see Price figures): the last price of a zone AWS stopped pricing (spotPriceStale), and the prices of usw1-az2 (legacyZone);
• then the rows with a spotPrice of 0: the zone had no spot price in its latest run (no spot there, or a failed fetch).

results is [] when the type has no price for the filters; a type that does not exist is a 404 ({"error": "unknown instance type: zz9.large"}).

Result fields

The other fields (instanceType, cpuManufacturer, vcpus, memoryGiB, gpus, cpuSpeedGhz, architecture, and the GPU, accelerator, storage, network and EBS fields) mean what they mean in Instance type fields.

curl -sG "https://go.runs-on.com/api/instances/m7a.xlarge" \
  -d zoneId=use1-az4 -d platform=linux
{
  "results": [
    {
      "timestamp": "2026-10-08T20:44:18Z",
      "region": "us-east-1",
      "az": "us-east-1b",
      "zoneId": "use1-az4",
      "instanceFamily": "m7a",
      "instanceType": "m7a.xlarge",
      "platform": "Linux/UNIX",
      "cpuModel": "AMD EPYC 9R14",
      "cpuManufacturer": "AMD",
      "vcpus": 4,
      "memoryGiB": 16,
      "gpus": 0,
      "cpuSpeed": 2875,
      "passmark": 2875,
      "cpuSpeedGhz": 3.7,
      "onDemandPrice": 0.23184,
      "spotPrice": 0.0941,
      "spotPriceAt": "2026-10-08T17:00:00Z",
      "spotPriceStale": false,
      "legacyZone": false,
      "architecture": "x86_64",
      "networkPerformanceGiBps": 12.5,
      "ebsMaxBandwidthMbps": 10000,
      "ebsMaxIOPS": 40000,
      "ebsBaselineBandwidthMbps": 1250,
      "ebsBaselineIOPS": 6000,
      "networkPerformanceGbps": 12.5,
      "networkBaselineGbps": 1.562,
      "networkBurstable": true
    }
  ]
}

Every zone of us-east-1 for t2.2xlarge: AWS stopped pricing use1-az3 in September, so its last price comes after the others.

curl -sG "https://go.runs-on.com/api/instances/t2.2xlarge" \
  -d region=us-east-1 -d platform=linux \
  | jq -c '.results[] | {zoneId, spotPrice, spotPriceAt, spotPriceStale}'
{"zoneId":"use1-az5","spotPrice":0.0881,"spotPriceAt":"2026-10-08T20:00:00Z","spotPriceStale":false}
{"zoneId":"use1-az6","spotPrice":0.1142,"spotPriceAt":"2026-10-08T17:00:00Z","spotPriceStale":false}
{"zoneId":"use1-az1","spotPrice":0.1155,"spotPriceAt":"2026-10-08T19:00:00Z","spotPriceStale":false}
{"zoneId":"use1-az4","spotPrice":0.1192,"spotPriceAt":"2026-10-08T12:00:00Z","spotPriceStale":false}
{"zoneId":"use1-az2","spotPrice":0.124,"spotPriceAt":"2026-10-08T16:00:00Z","spotPriceStale":false}
{"zoneId":"use1-az3","spotPrice":0.3712,"spotPriceAt":"2026-09-15T21:00:00Z","spotPriceStale":true}

CPU of a family#

GET /api/cpu-info

The processor and PassMark score of one family: family is required (400 without it), in lower case, and a family the site has no CPU record for is a 404.

Response fields

curl -sG "https://go.runs-on.com/api/cpu-info" -d family=c8i
{
  "cpuInfo": {
    "instanceFamily": "c8i",
    "cpuModel": "Intel(R) Xeon(R) 6975P-C (x86_64)",
    "cpuPerformance": 3256,
    "passmarkSource": "estimated",
    "passmarkSourceFamily": "m8i",
    "passmarkMeasuredAt": "2026-10-08T00:00:00Z",
    "passmarkBasis": "ec2-cpu",
    "recordedAt": "2026-10-08T20:44:18Z"
  }
}

Family exports#

Each family page, such as /instances/ec2/m8a, has its data in three other formats. Family names are lower case: another spelling, and a trailing slash, redirect (301) to the lower-case URL, and {family}.html to the page. The exports quote their page's figures, its PassMark score included, and may be cached for 5 minutes.

• GET /api/instances/{family}.json - JSON, the default without an extension. An unknown family is a 404 with close matches: {"error": "instance family 'zz9' not found", "close_matches": []}.
• GET /api/instances/{family}.md - Markdown, also served for Accept: text/markdown: specs and a table of the types, prices per platform in each type's cheapest region and in every region, and the regions. A retired family says so under its summary. An unknown family is a 404 in Markdown.
• GET /instances/ec2/{family}/regions.atom - an Atom feed with one entry per region the family was first priced in after the site started tracking regions, newest first, linking to the page's region list. Unknown families are a 404.

JSON fields

The JSON export uses snake_case names. Prices follow Price figures, per region.

An instance_type_specs entry:

A spec field is absent when it does not apply.

An accelerator:

A prices entry:

A pricing_time_series entry:

curl -s "https://go.runs-on.com/api/instances/m8a.json" \
  | jq '{cpu_model, cpu_clock_ghz, cpu_speed, passmark_source, regions, retired, large: .prices["Linux/UNIX"]["m8a.large"][0:2]}'
{
  "cpu_model": "AMD EPYC 9R45",
  "cpu_clock_ghz": 4.5,
  "cpu_speed": 3801,
  "passmark_source": "measured",
  "regions": ["ap-northeast-1", "ap-south-1", "eu-central-1", "eu-west-1", "us-east-1", "us-east-2", "us-west-2"],
  "retired": false,
  "large": [
    {"region": "us-east-2", "typical_spot_price": 0.0212, "on_demand_price": 0.12172},
    {"region": "ap-south-1", "typical_spot_price": 0.0247, "on_demand_price": 0.12806}
  ]
}

curl -s "https://go.runs-on.com/api/instances/m8a.md" | head -12
# m8a EC2 instance family

AMD EPYC 9R45 at 4.5 GHz, x86_64, 1 to 192 vCPUs, 4 to 768 GiB.

## Specs

- **Architecture:** x86_64
- **Processor:** AMD EPYC 9R45 (4.5 GHz sustained)
- **vCPUs:** 1 to 192
- **Memory:** 4 to 768 GiB
- **Single-thread PassMark:** 3801, measured on m8a, 2026-10-08: the median of its types in [runs-on.com benchmarks](https://runs-on.com/benchmarks/aws-ec2-instances/)
- **Platforms:** Linux/UNIX, Windows

In the Atom feed, an entry's title is the region, its updated the time the family was first priced there, its summary names the region's location, and its link points to the page's #available-regions. The regions of the day the site started recording the family (the first day of tracking, or the family's first price) are left out: that day says nothing of when AWS launched them. The feed is empty for a family that has gained no region since.

Sitemap and robots.txt#

GET /sitemap.xml

The home page, /finder, /instances/ranking, /compare, /api, and the page of every active family, with the day of its newest price as lastmod. A family is active while it has a price less than 15 days older than the site's newest price; a retired family keeps its page, marked retired and not indexed, but leaves the sitemap, the ranking and the finder.

curl -s "https://go.runs-on.com/sitemap.xml" | grep -c "<loc>"

GET /robots.txt

Crawlers may read every page, and of the JSON API only the two endpoints the finder and compare pages load their content from.

User-agent: *
Allow: /api/facets
Allow: /api/compare
Disallow: /api/

Sitemap: https://go.runs-on.com/sitemap.xml