OpenAI Releases GPT-6 Astra With Staged Access Across ChatGPT, API, and Cloud Partners

·4 views

OpenAI has launched GPT-6 Astra, described as its most capable model to date, with a phased rollout beginning with a limited group of organizations. Broader access is planned within days across ChatGPT Plus, Pro, Business, and Enterprise tiers, as well as the OpenAI API, Microsoft Azure, and AWS Bedrock. The model is built around extended reasoning, multi-agent coordination, and computer-use tasks — capabilities designed for complex, multi-step workflows rather than simple prompt-and-response interactions. Advanced cybersecurity features are being introduced more cautiously, initially restricted to trusted testers in OpenAI's Daybreak program before wider release. OpenAI has not yet published final API pricing, and availability of specific capabilities varies even where the underlying model is being rolled out.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds browser-native PNG compressor in vanilla JS, no libraries needed

A developer set out to build a fully client-side image compressor after noticing that most online tools upload sensitive files, such as passport scans, to remote servers. The project revealed that browsers can natively encode only three formats via canvas.toBlob() — PNG, JPEG, and WebP — and silently fall back to PNG when an unsupported format is requested, without throwing any error. Real PNG compression, the developer found, requires manually writing the file byte by byte, including building an optimised colour palette of up to 256 entries to replace full 24-bit colour data. Key implementation pitfalls included forgetting the per-scanline filter byte, which causes images to render diagonally skewed, and correctly packing bit depth to avoid wasting storage on images with few colours. The developer documented four such traps encountered during the build, aiming to save others the time lost debugging each one.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

InfoHUD Pro Lets Windows Users Pin Floating Info Panels on Their Desktop

A developer has launched InfoHUD Pro, a Windows 10/11 utility available on the Microsoft Store. The app allows users to create floating, always-on-top HUD panels displaying content such as text, spreadsheets, charts, images, PDFs, and web pages. Multiple HUDs can be open simultaneously, with options to move, resize, lock positions, or toggle visibility with a single click. The tool is designed with multi-monitor users in mind, helping keep reference material visible during work. The developer has indicated that updates are ongoing and is actively seeking user feedback.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Dev Stacks Gained Five Agent Layers Fast — but One Critical Layer Is Still Missing

Over the past two years, AI-assisted development introduced several new infrastructure layers, including agent tools like Cursor and Codex, context files, MCP-based tool integrations, isolation via git worktrees, and orchestration systems for managing multiple agents. Each of these layers was adopted quickly because its absence caused immediate, visible problems. However, the article identifies a sixth layer that most teams still lack: a coordination layer that checks whether multiple agents' plans can logically coexist before any code is written. Without it, two agents can each complete their tasks cleanly, pass all checks, and merge without conflict — yet still break the system by working from incompatible assumptions. The proposed solution is a declarative ledger where agents register semantic scopes of intended changes, allowing deterministic conflict detection at the planning stage rather than after the damage is done.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Kinde Webhooks Can Auto-Revoke AI Agent Access When a User Is Offboarded

A developer tested what happens when an AI agent built on Anthropic's Claude API continues running after its user is suspended mid-task in the Kinde identity platform. The experiment revealed that revoking a user in Kinde does not invalidate an already-issued OAuth access token, meaning the agent kept operating even after the user lost authorization. This is not a bug in Kinde or OAuth, but an inherent limitation of stateless, cryptographically signed tokens that verify locally without checking a central database. To close the gap, the developer built a system using Kinde webhooks to update an app-side user record the moment a suspension occurs, then enforced a status check before every agent tool call. The result is a practical pattern for ensuring offboarded users' AI agents are stopped in near-real time, regardless of token expiry.

0 comments Read more at DEV Community