The Real-SWE benchmark, which evaluates AI coding agents using proprietary real-world enterprise codebases, has been released, with Fable 5.1 ranking first with a 38.8% resolution rate.
Draw Things has launched the public beta of 'Local Code', supporting local LLM inference and code execution within a macOS sandbox environment.
NVIDIA has released 'PAIR', an open-source tool that connects multiple local computers to distribute AI inference requests.
The HacktronAI team used AI agents to gain access to OpenAI internal repositories, while the Claude Opus 5 model generated an exploit in an ASLR-enabled environment in just 3 hours, indicating a rapid improvement in AI models' vulnerability development capabilities.
Frontier labs seek to preserve premiums through regulation-based pacing.
Anthropic, Google, and OpenAI have launched their frontier AI models in separate public and vetted versions, with benchmarked models accessible only to identity-verified organizations.
OpenAI CEO Sam Altman stated that a 2026 IPO is inappropriate due to AI safety issues and a lack of social acceptability.
NVIDIA is providing billions of dollars in financial guarantees for large-scale projects such as the Ohio data center, earning it the moniker 'Central Bank of AI,' but concerns are raised about the boundary between demand creation and artificial demand generation.
Reverse engineering of the Apple Neural Engine reveals that the LLM inference bottleneck is data movement rather than compute, and it is integrated into the GPU in the M5 chip.
Dario Amodei proposed a three-step plan centered on introducing embedded evaluators, democratic coordination, and global coordination to regulate the pace of AI development.
Anthropic researchers released a mathematical framework proving that Induction Heads are the core mechanism for In-context Learning in 2-layer Attention-only Transformers.
Bipartisan US senators are negotiating a bill that imposes a duty of care on frontier AI developers and blocks the release of risky models.
Tencent has released AuK-Flash, a 1.5B parameter model that supports fast voice generation and editing via 4-step inference.
At COLING 2025, a framework and parallel corpus were released that use LLMs trained on small data to restore the endangered Nüshu language.
The open-source library Litelm has been released, stripping away unnecessary features from LiteLLM to achieve a lightweight footprint of 2,900 lines.
OpenAI analyzed ChatGPT conversations to detect a school attack plot and reported it to Argentine police through the FBI.
Catalyst has released a tool that measures and verifies key elements among 52 variables in compiled Rust models in just 1.37 seconds.
PR #27841 has been submitted to llama.cpp to add MMQ settings for AMD RDNA2 GPUs (MI50, MI60).
Anthropic released the 'Economic Scenario Explorer' to simulate changes in the US economy by 2030 based on AI adoption speed and capabilities.
The Agnes-3.0-Flash 33B model, released on Hugging Face, supports a hybrid attention architecture and a 262k context window.
The OpenAI Codex team released a guide to removing unnecessary instructions and managing context budgets in line with GPT-6 Astra's performance improvements.
Cognition released SWE-2, based on Kimi K3, achieving less than a 1 percentage point difference from Fable 5.1 on the FrontierCode benchmark while reducing costs by 64%.
NVIDIA has released Switchyard, an open-source tool that automatically routes LLM calls to the optimal model.
Perplexity uses GPT-6 Astra to automate code testing and production monitoring, reducing verification frequency.
AI Briefing