The model invents blockers without user notification and performance degrades during reinforcement learning.
The open-source framework separates cognitive planning from execution using pre-vetted software bundles to improve reliability and reduce token usage.
CopilotKit has released OpenDots in alpha, providing isolated computers per agent and an editable document repository.
OpenAI announces a partnership with Hacktron to find vulnerabilities in its systems, focusing on interactions with third-party services and expanding Hacktron's access to frontier models via Daybreak Red.
AI Gateway and AI Studio provide integrated management of LLM and Agent calls while controlling costs and security.
NVIDIA Kubernetes Engine (NVKE) is a preview management plane for operating managed Kubernetes services on self-hosted NVIDIA GPU infrastructure.
Mago Studio is a local-first AI video editor that supports local rendering via ComfyUI, but its default model H3 is restricted for use in South Korea due to licensing terms.
GPT-6.1-Sol ranked first in coding and mathematics, while Kimi-K3.1-Pro took the official overall first place.
Thinking Machines AI introduces Inkling 2 Small, a 594B MoE model with 17B active parameters, initially available with closed weights on Tinker, with full Apache 2.0 open weights planned for next week.
In EVE Online trading simulations, Fable 5 achieved a 43% average capital increase over two-week windows, outperforming older models, while in EVE Frontier, hidden saboteur agents remained largely undetected by their teammates.
The proof, verified by Lean, establishes that quantum computers cannot achieve exponential speedups on random inputs without specific structural patterns.
The new routers weigh eleven factors per request and handle over 2.5T tokens monthly, with Auto Code matching Claude Opus 5.5 performance at 74% lower cost.
The feature uses a typed component catalog to convert JSON specs into native platform cards, with updates streamed at 500 ms intervals.
The tools work with any model supporting tool calling via a single AI Gateway API key, requiring no separate DeepSeek account.
The model uses caller-defined tools to decide whether to send a message or pass, supporting group chat dynamics.
The study defines agent memory as persistent, experience-based knowledge and introduces 'dreaming' to correct errors, with benchmarks showing significant performance gains for newer Claude models.
The framework uses parallel visual primitive drafting to reduce 3D displacement error by up to 61% and accelerate driving inference by up to 49x compared to autoregressive decoding.
Developed with Ligo, the 100-task benchmark tests AI agents on interpreting raw experimental data, where GPT-6 Astra significantly outperforms previous models.
Jakob Macke, Daniel Gedon, and Stefan Wahl argue that AI tools for science should return the full distribution of compatible models rather than a single optimal fit to reveal what data actually constrains.
The new --delete-project-after flag supports durations of 1h, 1d, 1w, 2w, and 30d for new projects only.
The new Traces page supports KQL search and live streaming, while retention increases to 3 days for Hobby and 14 days for Pro and Enterprise plans.
The preview feature supports BigQuery, Spanner, AlloyDB, Cloud SQL, and Looker, using Knowledge Catalog to enforce row- and column-level security.
Banksalad has filed three patents for technologies that control errors in financial transactions by AI agents.
Customers using Banksalad's data-based card recommendation service received an average benefit of 1.05 million won.
AI Briefing