What we shipped, when.

KV: per-key TTL support.

• Added ttl option to KV.put() — sets an absolute expiry (in seconds) for the value.
• Added KV.expire(key, ttl) for changing the expiry of an existing key.
• Fixed a regression where compute-edge deploy --pops <subset> would still warm cold isolates at non-target PoPs.

Wellington PoP (wlg-01) is generally available.

• 14th PoP, covering NZ South traffic from Wellington data centre (Spark IX).
• Inter-PoP latency to akl-01 measured at ~5 ms; to syd-02 at ~28 ms.

If you've been routing NZ South traffic through akl-01, you'll see a ~30 ms reduction in p50 once Wellington picks it up automatically.

Python runtime in preview.

• Python 3.12 on the runtime via Pyodide-WASM. Handler shape mirrors the JS one (async def fetch(req): ...).
• Cold start typically 18−24 ms (slower than JS isolates but well below container starts).
• No native modules in preview — pure-Python or wasm-compiled wheels only.

Preview is open to all paid plans. Send feedback to [email protected].

Public status page; per-PoP latency in deploy output.

• Public status page at /status with 30-day per-PoP uptime.
• compute-edge deploy now reports per-PoP propagation time in the output (the 284 ms, 190 ms column).
• Logs viewer: timezone now respects the system locale instead of always UTC.

Spending caps; misc fixes.

• Per-account spending cap setting. Defaults to $0 on Hobby (so accidental usage spikes can't bill you).
• Fixed a CLI panic on Windows when the home directory contained a non-ASCII character.
• Fixed double-counting of cached requests in usage metrics.

Org plan; SSO (SAML).

• New Org tier — volume pricing, SSO/SAML, audit log, signed BAA/DPA.
• SCIM provisioning via Okta and Microsoft Entra ID.
• Audit log with 90-day retention on Pro, configurable retention on Org.

Object replication latency: down ~40%.

• Reworked the Object replication protocol to use the inter-PoP backbone instead of riding the public anycast network.
• p50 cross-PoP replication: 410 ms → 240 ms. p99: 2.8 s → 1.4 s.

Hong Kong PoP; HTTP/3 at the edge.

• hkg-01 brings APAC coverage to 6 PoPs.
• HTTP/3 (QUIC) enabled on all PoPs for inbound traffic. Outbound from functions stays HTTP/2 for now.
• Functions: streaming responses no longer time out at 30 s; the limit is now 5 min, gated by plan.

Rust runtime GA; cold starts < 5 ms.

• Rust runtime (WASM-compiled) is now generally available. Same handler shape as JS via the compute-edge-rs crate.
• JS cold starts measured down from ~12 ms to ~4 ms (V8 isolate snapshot warmup).
• Removed deprecated compute-edge functions list — use compute-edge status.

KV global eventual consistency.

• KV writes now replicate to every PoP region (was: same region only).
• Reads remain PoP-local. Convergence: typically 200−800 ms.

Public launch.

• Functions, KV, and Objects on 4 PoPs: tll-01, fra-01, iad-01, sin-01.
• Hobby tier free; Pro $9/dev/month at launch (raised to $12 in 2026-01 with grandfathering).

If you've been with us since the launch beta, thank you. We will not forget it.