Two new models with a big jump in agent capabilities and a practical piece on front-end performance in highly interactive screens.
GPT-6 Astra operates the computer in almost half the time of GPT-5.6 Sol
OpenAI released GPT-6 Astra, which according to the company is state of the art in computer use, browsing, software engineering, science and professional work. It fills out forms, updates CRM records, organizes calendars, researches and summarizes on the web and tests website features. In an OSWorld 2.0 simulation, it outperformed GPT-5.6 Sol in about 47% less time per task, and it is rolling out to ChatGPT, the API, Azure and AWS Bedrock.
Administrative tasks that require moving between several systems come closer to being fully delegated. Companies can map these routines and test agents with controlled permissions.
How Dexpert helps: Artificial IntelligenceSources
Gemini 3.8 Flash improves coding and agents at 3.7's price
Google released Gemini 3.8 Flash, its third Flash model in six weeks, with gains in software engineering, agentic tasks and multi-step reasoning, at the same price as 3.7 Flash: $0.75 per million input tokens and $3.75 output. According to the company, on DeepSWE v1.1, which measures autonomous solving of complex engineering problems, it beats most larger models at a fraction of the cost.
The release pace shows it pays to keep your own evaluations to switch models often and capture gains without higher costs. Architectures that isolate the model from the rest of the system make these switches easier.
How Dexpert helps: Artificial IntelligenceSources
Why the browser's main thread is the most expensive resource in a web interface
An article explains that, on highly interactive screens with live data, scrolling, animation and typing at once, the interface freezes whenever the browser's main thread is busy, however small the bundle or fast the network. Almost everything runs on it: the app's JavaScript, events, network responses, the framework itself and the steps to draw each frame. The post shows how to split work and move tasks off the main thread, for example by handing animations to the compositor.
Dashboards, operations systems and apps with real-time data run more smoothly when the front-end architecture accounts for the main thread from the start. Small freezes affect how users perceive product quality.
How Dexpert helps: Web DevelopmentSources
- kciter.so: The Browser's Main Thread Is Expensive
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.