All posts

iOS 27 arrives, Java 27 with G1 by default and a 4B model that speeds up Postgres queries

Curated by Dexpert, with the help of AI.

// Share this post

dexpert.com.br/en/blog/2026-09-21-ios-27-java-27-small-model-postgres-query-plans/

News on Apple's platforms, a new Java release and an experiment showing how small models can optimize databases.

iOS 27, iPadOS 27 and macOS 27 arrive with more Siri and visual intelligence

Apple released the new versions of its operating systems, including iOS 27, iPadOS 27 and macOS 27. New features include a Siri mode in the Camera, the new visual intelligence experience on iPhone, iPad, Mac and Vision Pro, Write with Siri, a standalone Siri app, automatic topic grouping of tabs in Safari and creating shortcuts from a text description.

Opportunity for your business

With more AI features in the system, users come to expect apps to work with Siri and shortcuts. Companies with iOS apps can test compatibility with the new versions and plan integrations.

How Dexpert helps: Mobile Apps

JDK 27 is generally available with G1 as the default collector and compact object headers

JDK 27, the reference implementation of Java 27, is now generally available with nine JEPs. They include G1 as the default garbage collector in all environments, compact object headers by default, which cut memory use, post-quantum hybrid key exchange in TLS 1.3 and in-process redaction of sensitive data in JFR. The release also brings hundreds of smaller enhancements and thousands of bug fixes.

Opportunity for your business

Java applications can use less memory and behave more predictably just by upgrading. Test enterprise systems on JDK 27 and plan migrations from older versions.

How Dexpert helps: Enterprise Apps & SaaS
Image: Rohan Bansal

A 4B model trained with reinforcement learning produces faster query plans than Postgres

Rohan Bansal showed that a small open model, the 4-billion-parameter Qwen, can be trained with fine-tuning and reinforcement learning to produce better query plans than Postgres's defaults. Since it is easy to measure whether a plan is good, just run it and time it, the problem fits this kind of training well. On 113 join-heavy queries, the model, which at first could not produce a plan for 99 of them, reached a 44.7% latency reduction.

Opportunity for your business

Small, specialized models can solve well-defined technical problems better than generic rules, at low cost. Tasks with easily measured results, such as query performance, are good candidates for this kind of solution.

How Dexpert helps: Backend & APIs

// Share this post

dexpert.com.br/en/blog/2026-09-21-ios-27-java-27-small-model-postgres-query-plans/

Curated by Dexpert, with the help of AI.

Want to put one of these to work in your company?

Dexpert builds custom systems, apps, integrations and AI solutions. Tell us what you need and get an estimate, or take the free digital maturity assessment.