Actix Web v4.13.0 boosts HTTP/2 upload throughput, raises MSRV to 1.88
Actix Web v4.13.0 raises the MSRV to Rust 1.88 and targets higher HTTP/2 upload throughput by increasing default flow-control windows and adding h2 tuning APIs.

Actix Web published v4.13.0 with two headline changes: "Minimum supported Rust version (MSRV) is now 1.88." and "Improve HTTP/2 upload throughput by increasing default flow control window sizes. #3638". The release, shown on GitHub as "actix-web: v4.13.0 Latest" and authored by JohnTitor, includes new HttpServer tuning methods named HttpServer::{h2_initial_window_size, h2_initial_connection_window_size} to let operators adjust HTTP/2 flow-control behavior.
The release artifacts on GitHub show the tag web-v4.13.0 with commit hash prefix 8b7270a and a commit marked "This commit was created on GitHub.com and signed with GitHub’s verified signature." The signature lists "GPG key ID: B5690EEEBB952194" and the page snippets include UI items like "Assets 3", "1 person reacted", and the label "Latest". The captured page also shows two timestamp snippets, "18 Feb 10:52" and "18 Feb 11:10", and an original report reference to "2026-02-21", so the release timing displays inconsistent date artifacts that should be reconciled on the live GitHub page before acting on a timeline.
The HTTP/2 work links directly to actix-http changes: actix-http: v3.12.0 also lists "Increase default HTTP/2 flow control window sizes. #3638" and "Expose configuration methods to improve upload throughput. #3638", indicating the change spans both libraries. The release notes do not list numeric default window sizes; the changelog instead exposes the new tuning API and recommends reviewing PR #3638 for exact values. Other developer-facing additions in actix-web v4.13.0 include "Add `HttpRequest::url_for_map` and `HttpRequest::url_for_iter` methods for named URL parameters. #3895" and "Add `experimental-introspection` feature to report configured routes #3594", while both actix-web and actix-http add a "Add config/method for `TCP_NODELAY`. #3918".
Stability fixes shipped alongside the features. actix-web lists "Fix panic when `NormalizePath` rewrites a scoped dynamic path before extraction (e.g., `scope("{tail:.}")` + `Path`). #3562" and "Do not compress 206 Partial Content responses. #3191". actix-http notes fixes for IO edge cases with "Fix truncated body ending without error when connection closed abnormally. #3067" and "Fix lingering sockets and client stalls when responding early to dropped chunked request payloads. #2972", reflecting a cross-repo push to address upload and connection robustness.

The release sits in a wider performance conversation in the community. Captured benchmark claims include "may minihttp = around 21.000 req/s", "actix = around 19.000 - 20.000 req/s", "axum = around 17.000 - 18.000 req/s", and "ntex neon-uring = around 23.000 - 25.000 req/s", alongside a community note that "Tokio is currently also experimenting with io uring implementation directly inside tokio library under tokio-unstable flag" and that io_uring "reduces syscalls and enables zero copy between user space and the kernel." The release record also preserves a cautionary perspective: "Benchmarks like this draw fire because although results are presented definitely, they can occasionally be misleading. Two languages/frameworks may have similar performance properties, but one of the two has sunk a lot more time into optimizing their benchmark implementation, allowing it to pull disproportionately ahead of its comrade. It tends to be less of an issue in the long run as benchmarks mature and all implementations get more optimized, but it’s a good idea to consider all benchmarks with a skeptic’s critical mind. [...] Fortunes is the most interesting of TechEmpower’s tests because it does more."
Operators and maintainers should note the MSRV bump in both actix-web v4.13.0 and actix-http v3.12.0 and evaluate the new HttpServer::{h2_initial_window_size, h2_initial_connection_window_size} APIs in staging. Because the changelog omits the numeric default window values, teams should review PR #3638 and the actix-http v3.12.0 diffs for exact settings and test upload-heavy workloads before upgrading production services. The release commit is signed and verifiable with GPG key ID B5690EEEBB952194, and the combined feature, tuning, and fix set makes v4.13.0 a pragmatic step toward better HTTP/2 upload behavior under real-world load.
Know something we missed? Have a correction or additional information?
Submit a Tip

