Upstash vs Valkey

What Upstash is

Upstash offers Redis and Kafka billed per request rather than per hour, so idle projects cost close to nothing. Per-request billing means a side project or a spiky workload costs nothing while idle, which per-hour pricing cannot match. Sustained high throughput is where that model turns against you and a provisioned instance wins.

What Valkey is

Valkey is a BSD-licensed, vendor-neutral in-memory data structure server descended from Redis OSS 7.2.4. It supports RESP2/RESP3 clients, replication, Sentinel, Cluster, transactions, scripting, Pub/Sub, streams and configurable RDB or AOF persistence. Compatibility is strongest with Redis OSS 7.2 and earlier; Redis Community Edition 7.4+ data files are not compatible. Secure deployment is not automatic and requires network isolation, access control and, where needed, TLS.

What Upstash is built to do

Per-request pricing

Bills by operation rather than by running hour.

Serverless Redis

Exposes a Redis-compatible store over HTTP for edge runtimes.

Global replication

Replicates data across regions for lower read latency.

What Valkey is built to do

Rich data structures

Stores strings, hashes, lists, sets, sorted sets, streams and specialized structures in memory.

Persistence choices

Supports periodic RDB snapshots, append-only logging, both together or no persistence.

Replication and failover

Uses asynchronous replication with Sentinel or Cluster deployment patterns.

RESP compatibility

Supports RESP2 and RESP3 and a documented migration path from Redis OSS through 7.2.

Choose Upstash if

• Serverless and edge apps that need Redis over HTTP with a free tier of 500K commands a month
• Low-traffic workloads billed per command: $0.20 per 100K commands with 1 GB free storage
• Teams that want predictable fixed plans from $10/month for 250 MB

Skip Upstash if

• You need an uptime SLA, encryption at rest or multi-zone HA — the Prod Pack is $200/month per database
• Your dataset exceeds 100 GB on pay-as-you-go
• You run a chatty workload where per-command billing beats a fixed plan only at low volume

Choose Valkey if

• Caches, ephemeral state, rate limits, queues, streams and real-time application data
• Redis OSS 7.2-or-earlier migrations validated against the actual client, modules and persistence files
• Teams that can operate protected standalone, Sentinel or Cluster deployments

Skip Valkey if

• The system requires disk-first durability or zero acknowledged-write-loss guarantees
• A Redis CE 7.4+ data file must be opened directly without a supported migration route
• The service would be internet-exposed or deployed without firewall, ACL and recovery controls

Upstash: pros & cons

• Per-request pricing suits spiky workloads
• Works well from serverless functions
• No servers to run
• Cost per request beats hourly only up to a point
• Latency depends on region choice
• Less control than self-managed Redis

Valkey: pros & cons

• Permissive open source: The server is community-developed under the BSD 3-Clause license.
• Familiar ecosystem: RESP clients and Redis OSS 7.2-era configurations, modules and data formats ease many migrations.
• Flexible roles: Data structures, expiry, scripting, transactions, Pub/Sub and streams cover cache and real-time patterns.
• Scale and availability options: Replication, Sentinel and Cluster support failover and sharding topologies.
• Compatibility has a boundary: Redis CE 7.4+ files are incompatible and future Valkey behavior can diverge.
• Unsafe defaults require hardening: Official guidance warns against direct internet exposure and calls for firewalling, binding, ACL/authentication and optional TLS.
• Durability is configurable: RDB, AOF and no-persistence modes have different data-loss, latency and recovery tradeoffs.
• Cluster consistency limits: Asynchronous replication allows windows of acknowledged-write loss, especially during minority partitions.