NVIDIA released four Cosmos 3 checkpoints for interactive simulation and multi-sensor driving scenes. Each has 16B trainable parameters, with commercial and non-commercial use permitted under OpenMDW1.1 terms.
Why it mattersTo use Sim-Auto for closed-loop interaction, builders need a simulation runtime and WSM renderer to update controls from driving inputs.
Google’s Gemini Omni 1.2 Flash is available through AI Gateway for text-to-video generation without a separate Google account.
Why it mattersTo use it, set the model name to google/gemini-omni-1.2-flash and use generateText or streamText in AI SDK to receive the video as a file.
OpenAI found that GPT-6 Astra completed about 10 percent of rubric items on safety-framed tasks versus about 16 percent on capabilities-framed tasks. Most of the performance gap persisted after researchers excluded explicit refusals and reasoning that stated a decision to withhold assistance.
Why it mattersIf your product uses AI coding agents, test equivalent tasks under different framings and check completion directly rather than relying on reported blockers.
ChatGPT Cyber would be available to the highest tier of Daybreak Red users. OpenAI said the model found 948 potential issues while scanning the Linux kernel and preparing patches.
Why it mattersProduct teams using open source software should budget time to review and apply patches as AI helps researchers uncover more vulnerabilities.
Gemini Omni 1.2 Flash adds native 15-second video generation and multi-turn editing. 720p video costs $0.07 per second, a 30% reduction from Gemini Omni 1.1.
Why it mattersDevelopers building video features can offer longer clips at lower generation costs and let users refine existing clips through follow-up edits.
OpenAI introduced GPT‑6 Cyber to improve vulnerability discovery and fixes while following researchers’ boundaries. GPT‑6 Cyber scored 48.5% on ExploitGym, compared with 32.1% for GPT‑5.6 Cyber.
Why it mattersProduct teams could use GPT‑6 Cyber to validate vulnerabilities and develop candidate patches, with human review for high-risk workflows.
Developers can automate embedding generation by choosing a model when setting up a vector search index. Built-in retries and error handling let embedding operations run without blocking primary database operations.
Why it mattersTeams building semantic search or RAG apps can spend less engineering time maintaining separate embedding pipelines.
Copilot’s integration with Xcode’s built-in coding assistant is in public preview starting with Xcode 27.3 beta. It supports all Copilot plans and requires macOS Tahoe 26.6 or later.
Why it mattersDevelopers building in Xcode can now choose between Copilot in the built-in coding assistant and the existing separate extension.
ChatGPT now makes GIFs from text prompts or uploaded photos. Users can change playback speed, edit frames, and share their GIFs.
Why it mattersProduct builders can use ChatGPT to prototype animated assets and adjust their timing and frames within the same tool.
Images 2.6 speeds up image generation by 50%. Developers can choose GPT‑Image‑2.6 Flare or GPT‑Image‑2.6 Sunburst in the API. The API gains image markup for targeted edits, while SVG exports support use across websites, apps, and print.
Why it mattersProduct builders can use image markup for precise revisions and adjust the new complexity parameter to speed up simpler editing tasks.
Ricursive released Ripple-65B Preview, pretrained on 12 trillion tokens and supporting up to 1M context length. The lab reports a TerminalBench 2.1 score of 80.52, claiming U.S. open-weight leadership in agentic coding. Developers can access the model through Hugging Face, Together AI, GMI Cloud, and OpenRouter.
Why it mattersTeams building coding agents can evaluate Ripple-65B Preview on their own tasks through several providers before committing to an integration.
OpenAI’s prototype analyzed 2,308 candidate Linux kernel findings and reproduced 1,724 vulnerabilities.
Why it mattersFor teams running products on Linux, keeping kernel patches current and isolating workloads can limit exposure to privilege escalation attacks.
