diff --git a/README.md b/README.md
index f5693f08..6731e4ad 100644
--- a/README.md
+++ b/README.md
@@ -186,7 +186,7 @@ is a summary; the full notes, with the reasoning behind each fix, are the record
 
 | Version | Published | What landed |
 |---|---|---|
-| **0.9.5** | Unreleased | **Feature:** Rust consumers can compile sampled pprof hooks out with `default-features = false` while retaining the allocator, profiler no-op stubs, and internal exact DHAT. |
+| **0.9.5** | 2026-08-27 | **Feature:** Rust consumers can compile sampled pprof hooks out with `default-features = false` while retaining the allocator, profiler no-op stubs, and internal exact DHAT. |
 | **0.9.4** | 2026-08-23 | **Fix:** replace the Rust-only ARM64 atomics allowlist with capability detection shared by every integration path. |
 | **0.9.3** | 2026-08-21 | **Fix:** the crate would not build for `aarch64-pc-windows-msvc`. mimalloc's MSVC atomics wrapper reaches for ARM64 `Interlocked` intrinsics that `clang-cl` does not declare, so a `cargo-xwin` cross build failed to compile ([#223](https://github.com/zackees/mimalloc-pprof/issues/223), [#224](https://github.com/zackees/mimalloc-pprof/pull/224)). **Also the first release carrying the v4 line**, merged into `main` after 0.9.2 ([#129](https://github.com/zackees/mimalloc-pprof/pull/129)): upstream engine pin `579f8c0e` -> `bcee5a88` ([#80](https://github.com/zackees/mimalloc-pprof/issues/80), [#113](https://github.com/zackees/mimalloc-pprof/pull/113)), the new experimental `mi_option_purge_zeroes` zero-tracking option ([#67](https://github.com/zackees/mimalloc-pprof/issues/67), [#79](https://github.com/zackees/mimalloc-pprof/pull/79)), the zeroing-`realloc` family exposed from Rust ([#83](https://github.com/zackees/mimalloc-pprof/issues/83), [#108](https://github.com/zackees/mimalloc-pprof/pull/108)), and several allocator fixes. |
 | **0.9.2** | 2026-08-01 | **Fix:** seeded sampling is now actually reproducible. `mi_prof_start_seeded` and `MIMALLOC_PROF_SEED` mixed the ASLR-randomised address of the thread's allocator state into the seed, so a pinned seed produced a different sample sequence in every process while the docs promised determinism ([#91](https://github.com/zackees/mimalloc-pprof/issues/91)). |