
Redis vs Valkey: performance, trade-offs and choosing for your project
Compare Redis and Valkey by benchmarks, licensing, features and migration costs.
Two related projects
Redis and Valkey both serve caches, sessions, shared counters and queues. Valkey originated from Redis 7.2 in 2024; the projects have developed separately since then. The right choice depends on the workload, required capabilities, operating environment and license.
Redis Open Source 8 and later offers RSALv2, SSPLv1 or AGPLv3. Redis 7.2 and earlier retain BSD-3-Clause; versions 7.4–7.8 use a different dual-license model. Valkey uses BSD-3-Clause. AGPLv3 is open source with copyleft conditions; BSD-3-Clause is permissive. Review the terms for the version and actual use. [1]
Features and compatibility
Both offer key–value structures, expiration, RDB/AOF persistence, replication and clustering mechanisms. Redis 8 integrates capabilities for search, JSON, time series and probabilistic data structures. Valkey’s available extensions depend on the release and modules. Its initial 7.2 release matched Redis 7.2.4’s protocol, API and data file formats, but newer commands, modules and migration paths should be checked individually. [2]
Benchmark: the winner changes with concurrency
A Lavik study from 21 October 2026 tested Redis 8.10.2 and Valkey 9.1.2 on a 16-vCPU VM with ten million 1-KiB values and pipeline 1. AOF and scheduled snapshots were disabled. Lavik develops a competing product; its methodology and results are published. [3]
| Operation / connections | Redis ops/s | Valkey ops/s | Redis p99.9 | Valkey p99.9 |
|---|---|---|---|---|
| GET / 1,280 | 976,087 | 913,756 | 7.039 ms | 7.743 ms |
| SET / 1,280 | 863,860 | 793,809 | 7.391 ms | 7.743 ms |
At 1,280 connections Redis delivered about 6.8% more GET throughput and 8.8% more SET throughput relative to Valkey. GET used 16 I/O threads for both; SET used 16 for Redis and 8 for Valkey, selected within the published methodology. These are therefore not identical thread configurations.
At 160 connections the ordering reversed: Valkey reached 459,449 GET/s versus 408,159 for Redis, and 453,171 SET/s versus 384,939. The p99.9 threshold captures the slow tail: approximately 99.9% of measured operations completed within it. The results do not predict performance in another application.
Advantages, limitations and migration
Redis has an established client ecosystem and integrated features that can simplify a design already using them. Its license options need evaluation, and reliance on Redis-specific commands or modules can make a later move harder. Valkey’s permissive license and Linux Foundation governance are attractive for a basic cache or shared sessions; verify extensions and managed-service availability before deciding.
For a disposable cache, migration may involve starting a new empty instance and warming it gradually. Check clients, TLS, authentication, timeouts and the resulting database load. Sessions and queues need a cutover plan for concurrent writes and a tested rollback. Verify command, module and backup compatibility in both directions for the versions in use. Replicas do not replace backups.
Which should you choose for the job?
| Scenario | Starting point | What to check |
|---|---|---|
| New catalog or API cache | Test both; consider Valkey when BSD matters | Measure latency, memory and managed-service price. |
| Existing Redis search, JSON or time series | Evaluate Redis and specific alternative modules | Check command and index compatibility and migration effort. |
| Sessions, rate limits or leaderboards | Both are candidates | Check client support, failure handling and consequences of lost state. |
| Stable existing Redis deployment | Compare the cost of staying and migrating | Migration requires compatibility tests, cache warm-up and an operational plan. |
| High throughput | Benchmark both for your workload | Compare p99, throughput and cost under the same SLO. |
These are starting points, not compatibility guarantees. A small single-instance application may get enough benefit from a local cache or a faster database query; shared sessions and global limits still need an appropriate shared solution.
Which should you choose for the job?
Section choice-by-scenario screenshotMeasure the complete request
User → API → Redis or Valkey → API response
└─ miss → primary database → cacheMeasure end-to-end API latency as well as GET: network, serialization, cache misses and database work can dominate. Test typical and peak traffic, then simulate failure. Record p95/p99, miss rate, CPU, RAM, errors and recovery time. A synthetic GET advantage may not help when most requests miss the cache and wait for an expensive SQL query. These are analytical examples, not measurements we performed.
Run a useful comparison
Fix the versions, server class, resource limits, network and persistence settings. Use representative data sizes, key distribution, GET/SET mix, concurrency and pipeline settings; document any necessary configuration differences. Warm up and repeat the test, recording throughput, p50/p95/p99, CPU, memory and errors. Confirm the load generator is not the bottleneck. Finally test recovery and end-to-end application latency. Avoid load testing production without coordination. [4]
For a new basic cache, Valkey is a strong candidate if BSD-3-Clause fits your requirements. Redis is compelling when its integrated features or existing deployment matter. For a stable system, compare migration cost with its expected benefits. Neither benchmark point establishes a universal winner.
Sources
Related Articles

Redis: how it works and when it makes an application faster
Learn how Redis works through caching examples, data structures, benchmarks and practical trade-offs.

Valkey: how the key–value store works and where to use it
An introduction to Valkey, shared state, caching, performance and migration from Redis.