AWS Cost
What AWS will charge, and how to pay less for it: a Lambda cost calculator, and which Savings Plan or Reserved Instance to buy for EC2.
Estimate the monthly AWS Lambda bill for one function from how often it runs, how much memory it has and how long it takes — requests, compute (GB-seconds, tiered) and ephemeral storage, with or without the free tier.
128–10,240 MB · 0.29 vCPU
Billed per 1 ms, rounded up — 100 ms each
512–10,240 MB; the first 512 MB is free
1,000,000 requests + 400,000 GB-s a month, per account
Instructions
Estimates the on-demand monthly cost of a single AWS Lambda function from its traffic, memory, duration and /tmp size. Everything is calculated in the browser; nothing is sent to AWS.
- Pick the Architecture — Arm (Graviton) is about 20% cheaper per GB-second.
- Enter how many Invocations the function gets and choose per month, day, hour or second.
- Set Memory and the Average duration from CloudWatch (the
Durationmetric, or theREPORTline in the logs). - Raise Ephemeral storage only if you configured more than the default 512 MB of
/tmp. - Read the total and the breakdown underneath; it updates as you type.
Tips
- Cost = requests × $0.20 per 1M + GB-seconds × the price per GB-second, where GB-seconds = invocations × duration (s) × memory (GB). Prices are the US East (N. Virginia) list prices from the AWS Lambda pricing page; most regions match, a few charge more.
- Duration price tiers apply to the whole account's monthly GB-seconds per architecture and region, not per function — with other busy functions you may already be in a cheaper tier.
- The free tier (1M requests + 400,000 GB-s a month) is shared by every function in the account, so untick it if other functions already use it up.
- More memory also means more CPU (1,769 MB ≈ 1 vCPU), so a function often gets faster as memory goes up — the cheapest setting is not always the smallest.
- Not included: provisioned concurrency, SnapStart, Lambda@Edge, data transfer, and the services around the function (API Gateway, CloudWatch Logs, SQS…).
- A Compute Savings Plan takes about 12% (1 yr) to 17% (3 yr) off duration and provisioned concurrency, never off requests. It is the only plan that covers Lambda; see the Savings Plans tab.
- Your inputs are kept in this browser and restored on reload.
- Hit Example and click one of the highlighted Presets — API backend, Cron job, Image resize or Stream consumer (which is big enough to reach the cheaper tiers).
Which commitment to buy for EC2 in Sydney (ap-southeast-2): a Compute Savings Plan, an EC2 Instance Savings Plan, a Standard RI or a Convertible RI. Linux, shared tenancy, t4g or t3a. Answer the questions and the page picks the cheapest plan your answers allow, and says why each of the others is not it. Prices are USD from the AWS Price List API, pulled 2026-10-05. The full write-up is the fa-doc AWS Savings Plans vs Reserved Instances for EC2.
Help me choose
Then the term, payment and amount
- 37% off On-Demand (1 yr, No Upfront), the biggest discount of the four.
- Covers any t4g size, OS or tenancy in Sydney by itself. When you scale up, add a second plan for the difference.
- ⚠️ Commit the Savings Plan rate, not the On-Demand price ($0.0106/h per instance). That would over-commit by about 58%, wasted every hour.
Why not the others
| Plan | Verdict | Why |
|---|---|---|
| Standard RI | Allowed, but worse |
|
| Compute Savings Plan | Allowed, but worse |
|
| Convertible RI | Allowed, but worse |
|
The four options
| Option | You commit to | Where it applies | Best for |
|---|---|---|---|
| EC2 Instance Savings Plan | $X per hour of spend, at Savings Plan rates | One family in one region (e.g. t4g in Sydney), any size, OS or tenancy | Staying on one family in Sydney for the whole term. The default pick. |
| Standard RI | N instances of one type | That type in one region. Size can flex within the family; the family never changes. | A zonal capacity reservation, or a way out through the RI Marketplace (needs a US bank). |
| Compute Savings Plan | $X per hour of spend, at Savings Plan rates | Any EC2 family, size, OS, tenancy or region, plus Fargate and Lambda | Maybe moving family (t4g ↔ t3a), changing region, or adding Lambda / Fargate. |
| Convertible RI | N instances of one type | That type in one region. Family, size, OS and tenancy change only by a manual exchange. | Rarely. A Compute SP gives the same discount without manual exchanges. |
There are really only two discount levels
In the Sydney price files each Savings Plan has the same rate as one RI type, to within rounding. AWS advertises “up to 66%” and “up to 72%”; t4g and t3a in Sydney reach about 50% and 62%.
| Discount level | Products | 1 yr | 3 yr |
|---|---|---|---|
| Flexible | Compute SP = Convertible RI | 22% – 27% | 44% – 50% |
| Locked to one family | EC2 Instance SP = Standard RI | 37% – 41% | 57% – 62% |
No Upfront – All Upfront, t-family.
Compare prices
One instance, Linux, shared tenancy, USD list prices from the AWS Price List API, pulled 2026-10-08. The % is how much cheaper than On-Demand.
| 1 yr · No Upfront | 3 yr · No Upfront | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sydney | t4g.micro | 2 | 1 GiB | arm64 | $7.74 | $4.89(37%) | $4.89(37%) | $6.06(22%) | $6.06(22%) | $3.36(57%) | $3.36(57%) | $4.31(44%) | $4.31(44%) |
| Sydney | t4g.small | 2 | 2 GiB | arm64 | $15.48 | $9.78(37%) | $9.78(37%) | $12.05(22%) | $12.05(22%) | $6.72(57%) | $6.72(57%) | $8.61(44%) | $8.61(44%) |
| Sydney | t4g.medium | 2 | 4 GiB | arm64 | $30.95 | $19.49(37%) | $19.49(37%) | $24.16(22%) | $24.16(22%) | $13.43(57%) | $13.43(57%) | $17.23(44%) | $17.23(44%) |
| Sydney | t3a.micro | 2 | 1 GiB | x86_64 | $8.69 | $5.48(37%) | $5.48(37%) | $6.79(22%) | $6.79(22%) | $3.72(57%) | $3.72(57%) | $4.82(45%) | $4.82(45%) |
| Sydney | t3a.small | 2 | 2 GiB | x86_64 | $17.37 | $10.95(37%) | $10.95(37%) | $13.58(22%) | $13.58(22%) | $7.52(57%) | $7.52(57%) | $9.64(45%) | $9.64(45%) |
| Sydney | t3a.medium | 2 | 4 GiB | x86_64 | $34.68 | $21.83(37%) | $21.83(37%) | $27.08(22%) | $27.08(22%) | $14.97(57%) | $14.97(57%) | $19.35(44%) | $19.35(44%) |
- Green is the cheapest commitment in the row for each term (several when they tie to the cent). “—” means AWS doesn't sell that plan for that instance there.
- Partial and All Upfront are shown as an effective rate: the upfront fee spread over the term. A total is everything paid over the term (12 or 36 months of 730 h).
- The highlighted row is the instance picked above, in Sydney.
Break-even: lock to t4g, then move to t3a
One t4g.micro, 1 yr No Upfront, total for the term. On the locked plan the commitment is wasted after the move and the t3a.micro is billed On-Demand. The Compute SP keeps covering what it can of the t3a.micro.
| Move to t3a.micro after… | EC2 Instance SP (t4g) | Compute SP |
|---|---|---|
| never | $58.68 | $72.72 |
| month 11 (90%) | $69.11 | $73.84 |
| month 10 (85%) | $74.32 | $74.40 |
| month 6 (50%) | $110.82 | $78.33 |
| month 3 (25%) | $136.89 | $81.13 |
The EC2 Instance SP only wins if you'd move in roughly the last 15% of the term, or never.
What none of the four discount
Burstable CPU credits. t4g, t3a and t3 launch in unlimited mode; CPU above baseline, averaged over 24 h, is billed as surplus credits. The Sydney Savings Plans price file has no rate at all for CPUCredits.
Public IPv4 address, billed per hour on its own line. At micro size that's a big share of the bill.
EBS volumes, snapshots and data transfer. And Spot: Savings Plans don't apply to it.
Instructions
Picks which commitment to buy for EC2 in Sydney — Compute Savings Plan, EC2 Instance Savings Plan, Standard RI or Convertible RI — and how much to commit. Everything is worked out in the browser; nothing is sent to AWS.
- Answer the five questions in Help me choose. Each answer can rule plans out.
- Pick the term and payment, and the instances that run 24/7 today.
- Read the result: the plan to buy, the $/h to commit (Savings Plans) or the RIs to buy, the monthly cost against On-Demand, and why each other plan is not it.
Tips
- The order of preference is EC2 Instance SP, Standard RI, Compute SP, Convertible RI: the cheapest plan still standing wins.
- If every plan is ruled out, two answers conflict. For example “might leave Sydney” and “need zonal capacity” leave nothing. Decide which need matters more, or stay On-Demand.
- A Savings Plan commitment is spend at Savings Plan rates, used up hour by hour. Unused commitment in an hour is lost, and usage above it is On-Demand. Plans stack, so add a second one when you scale up.
- A Savings Plan can be returned only within 7 days of buying it, in the same calendar month, if its commitment is $100/h or less. After that you are locked in.
- After a new instance has run for a week or two, check AWS's number in Billing → Savings Plans → Recommendations before you buy.
- The questionnaire uses No Upfront list prices for Sydney, Linux, shared tenancy. Use Compare prices for other regions, instances and payment options. Your answers and filters are kept in this browser and restored on reload.
- Hit Example and click one of the highlighted Scenarios. Each is one result from the fa-doc, for one t4g.micro.