Linode (Akamai) Specs (2026): Seven Plan Classes and What the API Publishes

1. The seven classes

Every instance in the catalogue belongs to one of seven classes, and the class is what tells you which resource is being sold. Counts and price ranges below are from the types API in September 2026; the US region column is from the availability API and is explained in section 5.

Forty-five of the 75 types carry a price in the API, and 30 do not: all 13 GPU types return a null monthly figure, as do the 17 g8 dedicated types. The dedicated price range in that table therefore comes from the g6 and g7 generations, which is the part of the line a buyer can price from the API alone.

2. The Shared CPU ladder

Ten types, from the Nanode to the 192 GB instance. Prices, cores, memory, disk and the two network figures are exactly as the API returns them.

The ladder is close to linear on memory and better than linear on disk: doubling from 32 GB to 64 GB doubles the price and the disk, and the transfer allowance stops growing above 64 GB, where four plans share the same 20 TB. The 2 GB entry at $12 is the size the storage page uses to compare what providers publish about their disks.

3. High Memory, Premium CPU and dedicated CPU

High Memory buys memory and gives up everything else: 48 GB arrives on 2 vCPUs with a 40 GB disk, so it is a cache or an in-memory index rather than a database host. The 32 GB page uses the 48 GB entry as the cheapest way to hold a large working set on this provider.

Premium CPU is the shared shape on newer processors, at roughly 1.8 times the price for the same cores, memory and disk: $43 against $24 at 4 GB, $346 against $192 at 32 GB. The API does not name the processor in either class, so what "premium" buys is stated by class rather than by model.

Dedicated CPU reserves the cores for one instance, at double the cores for the same memory: 32 GB on 16 dedicated cores at $288 against 32 GB on 8 shared cores at $192. The class also carries a g7 generation, and in the API it is the Premium ladder above under a second label: g7-dedicated-32-16 and g7-premium-16 are both $346 for 16 vCPUs, 32 GB, 640 GB and a 7,000 Mbit/s network figure, and the two ladders match plan for plan down to the transfer allowance. The g8 generation is listed without prices. That is the line to price against a bare-metal server, which the dedicated CPU page and the cloud versus bare metal comparison both do.

4. Network speed, which Linode publishes per plan

Most providers on this site state a port speed for a product line, or nothing at all. Akamai's API carries a network_out figure for every instance type, and it is a per-plan number rather than a line-wide claim: 1,000 on the Nanode, then 2,000, 4,000, 5,000, 6,000, 7,000, 9,000, 10,000, 11,000 and 12,000 up the shared ladder, 4,000 to 11,000 on dedicated CPU, and 16,000 on every GPU and Accelerated type. The unit is Mbit/s, so the range is 1 to 16 Gbit/s.

Two cautions in reading it. The figure describes the instance's network allowance and the API does not say what it is shared with, which is why the network page treats published port speeds as a ceiling rather than a promise. And the transfer allowance in the table above is a separate limit: the shared ladder keeps growing its network figure past 64 GB while the transfer allowance stays at 20 TB.

5. Where each class is available

The availability API returns 2,475 pairs, one for every combination of 33 regions and 75 types, and 750 of those pairs are in the ten US regions; 389 of the 750 are available. Counting the types available in each region gives a picture that the price tables alone do not show.

Read by class instead of by region, the split is wider. The Nanode and all nine Shared CPU types are in every one of the ten US regions. High Memory is in nine. Premium CPU is in six. Dedicated CPU ranges from three regions for some g8 generations to ten for the g6 types. GPU is in two or fewer, and the Accelerated class is in one. So the plan a buyer picks from the cheap end of the price table can be ordered almost anywhere, while a specialised instance may mean taking the region that has it rather than the region nearest the users. The US datacenter guide maps which city suits which coast.

6. The hourly rate and the monthly cap

Every type with a price carries two: an hourly rate and a monthly one. The relationship holds across the 45 priced types and the tables above show it in full: the monthly figure is the hourly rate multiplied by 666.67 on 43 of them, and the two exceptions are Accelerated instances where the printed hourly rate is rounded to two decimals ($0.53 against $352, $0.73 against $488). On the Nanode, $0.0075 an hour is $5.00 a month. On Linode 32GB, $0.288 is $192. On the 256 GB dedicated instance, $3.456 is $2,304.

666.67 hours is 27.8 days, which means the printed monthly price is a ceiling on what a month can cost rather than 30 days of the hourly rate. The billing documentation was not read for this page, so how the cap is applied at the end of a month is not stated here. What the API does establish is that both numbers are published for every plan, so a reader comparing an instance that will run for a week against one that will run for a year can do that arithmetic from the API alone.

7. What Linode does not publish

• A processor model. The API gives a class name and, for Premium, the word premium; the processor behind it is not named.
• Disk IOPS or throughput. Disk size is published; its speed is not.
• Prices for the g8 dedicated generation (17 of the 33 dedicated types) or for any of the 13 GPU types. Both return a null monthly figure in the API.
• What the per-plan network figure is shared with. network_out is a per-instance number with no host-level context.
• The billing cap's mechanics, on the API pages read for this one.

8. Measuring the parts that are not published

The processor, the disk and the achievable throughput are all measurable on the instance itself, and the method is the same one used across this site: lscpu for the CPU model, sysbench cpu --threads=1 --time=60 run for single-thread speed, a 4K fio run for the disk, and iperf3 against a machine near your users for the network. The benchmarks hub has the commands and the reasoning, the storage page compares what other providers publish, and the Linode review covers how the classes compare with the rest of this site.

Hourly billing is what makes that cheap here rather than a month-long commitment: an instance can be created, measured and deleted inside an afternoon, and the only figure the API does not give you in advance is the one you are trying to measure.