140 wires and counting

$ follow Google

Keep up with Google in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-10-03
stories 69

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

JAX adds type-2 and type-3 Hermitian eigensolves, lexsort batching cuts unique compile time

By RepoJournal · Filed · About Google · Composed from the cited sources · methodology

JAX extended generalized Hermitian eigensolvers and fixed a compile-time blowup in jnp.unique, while the Google client libraries shipped a python-genai minor release and three api_core transport changes.

Support type-2 and type-3 generalized Hermitian eigenproblems google/jax

by jabrailkhalil

jax.scipy.linalg.eigh(a, b) used to reject type=2 and type=3; it now solves A B v = w v and B A v = w v via a Cholesky reduction, with the reduced matrix Lᴴ A L and eigenvectors transformed back by L⁻ᴴ for type 2 and L for type 3. Existing calls are unaffected; the new branches are the ones previously raising.

Sort lexsort keys in batches so compile time doesn't grow with the number of keys google/jax

by bhaskargurram-ai

For jnp.unique(x, axis=...), _unique_sorted_mask ordered slices with a single lexsort carrying one key per column, and the generated comparator grew with the key count. Keys are now sorted in batches, so compile time no longer scales with the number of keys; in the reported case of jnp.unique(jnp.zeros((2, 1000)), axis=0), nearly all of the roughly 15 seconds on CPU was compilation.

[BugFix] Replace phase-2 PartialReduce with Pad-Reshape-Reduce in SPMD. google/jax

by Felix Chern

When ApproxTopK is partitioned across the reduction dimension, the second reduction step previously emitted a secondary PartialReduce that could drop elements from trailing shards. It is now a Pad-Reshape-Reduce, which is where to look if you were seeing truncated results under SPMD partitioning.

v2.28.0 googleapis/python-genai

by release-please[bot]

python-genai 2.28.0 adds lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to the Interactions Model enum and supports continuation_token in Interactions, following the separate feature commit.

feat(auth): Support sync credentials in AsyncAuthorizedSession googleapis/google-cloud-python

by daniel-sanche

AsyncAuthorizedSession now accepts standard credential types by wrapping them in an adapter, and the resumable upload path is being refactored to work with both aiohttp.ClientSession and AsyncAuthorizedSession instead of the former only.

Quick answers

What shipped in Google on October 3, 2026?
JAX extended generalized Hermitian eigensolvers and fixed a compile-time blowup in jnp.unique, while the Google client libraries shipped a python-genai minor release and three api_core transport changes. In total, 36 commits, 30 pull requests, and 3 releases landed.
Who contributed to Google on October 3, 2026?
9 developers shipped this update, including jabrailkhalil, bhaskargurram-ai, Felix Chern, copybara-service, mattjj, release-please[bot], Mark Daoust, and release-please, and 1 more.
What were the notable Google updates?
Support type-2 and type-3 generalized Hermitian eigenproblems, Sort lexsort keys in batches so compile time doesn't grow with the number of keys, and [BugFix] Replace phase-2 PartialReduce with Pad-Reshape-Reduce in SPMD.