Latest

Claude Code reported 91% refactor accuracy on a 150K-line codebase. And that single figure is both the best case for AI code refactoring tools in 2026 and the best reason to pause before you point one at production. The short version for web developers: run Claude Code CLI from the

Prompt optimization frameworks have their proof point and it is not subtle: OPRO beat human-designed prompts by up to 50% on Big-Bench Hard. Up to 8% on GSM8K. No person touched the instruction. Result comes straight from the Optimization by PROmpting paper, and that one line is the whole pitch.

GLM-5.3-Flash spent the back half of August 2026 answering to a name that wasn't its own. And if you were one of the developers hammering the mystery model called "Ox Alpha," arguing about its behavior on OpenCode and OpenRouter, you were beta-testing Z.ai'

Four steps. That's all ETGPO needs. Error collection, error taxonomy creation, error category selection, guidance generation. The whole method, spelled out in a paper sitting at arXiv 2602.00997. And that count tells you something. Nobody builds a taxonomy unless failures are piling up faster than fixes. Collect

Amazon has five names on an arXiv paper, identifier 2609.04184. And its title states the whole thesis: "Toward Frontier-Quality Declarative UI Generation at Small-Model Cost." Declarative UI generation means a model hands your frontend a structured description of an interface, a card or a form or a

A 2025 arXiv paper scaled a latent reasoning model to 3.5 billion parameters and 800 billion training tokens. And its performance on reasoning benchmarks kept improving, sometimes dramatically, up to a computation load equivalent to 50 billion parameters, without writing out a single visible reasoning step. That's