AWS Lambda Web Functions vs Lambda Web Adapter: cost
Researched 2026-10-13. Sources: announcement, pricing, FAQ.
What it is
Section titled “What it is”A new Lambda type that runs a normal HTTP server (Express, Hono, Fastify, Next.js) unchanged. The app listens on a port, and Lambda handles scaling, the HTTPS endpoint and multi-region routing.
- Up to 128 concurrent requests per instance. Classic Lambda handles 1.
- Billed for active CPU time + memory for the instance’s lifetime, not wall-clock time. Waiting on I/O costs no CPU.
--regionsflag replicates to several regions behind one global URL with latency-based routing. These are regions, not edge locations.- Immutable revisions with weighted endpoints for canary deploys.
- Scales to zero when idle (FAQ). Memory is billed “including time between requests” while an instance is alive. How long an idle instance stays alive isn’t documented.
- At launch: Node.js 24 on Arm only. Regions: us-east-1, us-east-2, us-west-2, Tokyo, Ireland.
vs Lambda Web Adapter (LWA)
Section titled “vs Lambda Web Adapter (LWA)”LWA is an open-source extension that runs the same web server on classic Lambda, with classic billing.
| LWA on classic Lambda | Web Functions | |
|---|---|---|
| Requests per instance | 1 | up to 128 |
| Billed on | wall-clock time × memory | active CPU + memory while the instance exists |
| Free tier | 1M requests + 400k GB-seconds/month | none found |
| HTTPS endpoint | Function URL (free) or API Gateway (+$1/M) | included |
| Multi-region | build it yourself | built in |
| Runtimes | any | Node 24 on Arm only |
Prices (us-east-1, Arm)
Section titled “Prices (us-east-1, Arm)”| Web Functions | Classic Lambda | |
|---|---|---|
| Compute | $0.0000276944 per vCPU-second of active CPU | $0.0000133334 per GB-second |
| Memory | $0.0000036667 per GB-second | included |
| Requests | $0.30 per million | $0.20 per million |
Free tier: I didn’t find one for Web Functions on the pricing page, the FAQ or in a web search. The classic free tier is only listed under Event Functions. The pricing examples say they “exclude the free tier”, but the page never says what that free tier is. Confirm with AWS support or check the bill after a test month.
Monthly cost, 1M–10M requests
Section titled “Monthly cost, 1M–10M requests”Assumptions: 512MB, 200ms per request, 20ms of it real CPU.
| Requests/mo | Avg concurrency | LWA (list price) | LWA (with free tier) | Web Fn (scales to zero) | Web Fn (1 instance always on) |
|---|---|---|---|---|---|
| 1M | 0.08 | $1.53 | $0.00 | $1.22 | $5.61 |
| 2M | 0.15 | $3.07 | $0.20 | $2.44 | $6.46 |
| 3M | 0.23 | $4.60 | $0.40 | $3.66 | $7.31 |
| 4M | 0.31 | $6.13 | $0.60 | $4.88 | $8.17 |
| 5M | 0.39 | $7.67 | $2.13 | $6.10 | $9.02 |
| 6M | 0.46 | $9.20 | $3.67 | $7.32 | $9.88 |
| 7M | 0.54 | $10.73 | $5.20 | $8.54 | $10.73 |
| 8M | 0.62 | $12.27 | $6.73 | $9.76 | $11.58 |
| 9M | 0.69 | $13.80 | $8.27 | $10.99 | $12.44 |
| 10M | 0.77 | $15.33 | $9.80 | $12.21 | $13.29 |
Real Web Functions cost falls between its two columns, closer to “scales to zero”. That column is a worst case: memory billed for every request’s full 200ms with no sharing.
Verdict
Section titled “Verdict”- Up to 10M requests/month, LWA with the free tier is cheapest.
- Without the free tier, Web Functions is about 20% cheaper ($1.22 vs $1.53 per million requests).
- At high concurrency (around 50M+ requests/month or bursty traffic), Web Functions gets close to 45% cheaper. Memory is shared across up to 128 requests, which leaves about $0.85/M in CPU and request charges, vs $1.53/M on LWA.
- I/O-heavy apps save more. CPU-heavy apps are cheaper on LWA, because the CPU rate is about 2× the GB-second rate.
- At small scale the reasons to pick Web Functions are fewer cold starts, a built-in multi-region URL and canary deploys. Not cost.
- If AWS turns out to apply the classic free tier to Web Functions, it would probably be cheapest across the whole range.