GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents
GPT-6 Cyber delivers major gains on key cybersecurity benchmarks over GPT-5.6 Cyber, including a 51% relative improvement on ExploitGym and a step function improvement when tested on staying within scope.
It is already helping to secure real-world software, uncovering and validating thousands of exploits in security-critical codebases such as Linux and Firefox.
Ant Group: Introducing Ling-3.2-flash, Ant Ling's native multimodal model: 560B total / 25B active parameters.
Introducing Ling-3.2-flash, Ant Ling's native multimodal model: 560B total / 25B active parameters.
Stronger complex visual understanding, long-horizon multimodal agents, 3D understanding & interaction, and medical intelligence. Outscores Opus 5 (max) in release evaluations: - Posted 2026-11-05.
ByteDance Seed publishes Qwen3-30B-A3B-MoE-SAE model repository
ByteDance Seed publishes Qwen3-30B-A3B-MoE-SAE model repository.
The Hugging Face page documents the ByteDance Seed update.
05Updates
Thinking Machines: Introducing Inkling 2 Small.
Introducing Inkling 2 Small.
Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor. Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
Docker and ActiveState Extending Secure Defaults to the Packages Developers Install
Docker Hardened Images (DHI) are open source under Apache 2.0, with a catalog of 4,000+ minimal images (and growing) built from source and continuously maintained by Docker.
The catalog also carries customizable Helm charts and over 65,000 hardened system packages.
arXiv: From Twisted Facts to Rogue Agents: Can LLMs Reliably Detect Multi-Dimensional Hallucinations?
Large language models (LLMs) are widely used for information access, content production, and agent based task execution, but their outputs remain vulnerable to subtle, high risk hallucinations.
The arXiv Language-Model Research page documents the arXiv update.
10Updates
What's new in Microsoft Foundry | October 2026
We introduced the Microsoft Copilot harness in public preview, made Browser Automation generally available, and brought the Decisions API to Foundry.
The managed harness is preview; long-running resilience for hosted agents is GA.
arXiv: Agent-R$^2$: Scaling Agents for Realistic Productivity Work via Real-Source Data Synthesis
A key bottleneck is the lack of scalable post-training data that reflects real work intent, authentic file environments, and deliverable-level evaluation.
The arXiv Language-Model Research page documents the arXiv update.
ReproBridge: Building Reproducibility into ML Research Artifacts
Reproducing machine learning experiments requires reconstructing procedures from information scattered across papers, released code, and supporting artifacts.
The arXiv Coding-Agent Research page documents the arXiv update.
SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization
LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods.
The arXiv Coding-Agent Research page documents the arXiv update.
Run Qwen3.8-27B locally, connect it to GitHub Copilot, and get a responsive coding experience with predictable inference costs, greater privacy, and the freedom to keep working offline.
Qwen3.8-27B is its primary optimization target because it pairs strong coding capabilities with a size suited to high-performance local deployment.
The Two-Ledger Model: Transaction-Grain Cost Accounting for Multi-Agent AI Systems Across Clouds, Models, and Systems of Record
The FinOps Open Cost and Usage Specification (FOCUS) gives billing data one schema across providers but remains settlement-grain even as it adds AI token and requester attribution.
The arXiv coding-agent research feed release records 2611.06989v1 for arXiv.
arXiv: Accepted but Wrong: Silent Failures of LLM Code Repairs after API Migrations
After a runtime or API migration, an LLM repaired procedure can run cleanly yet return wrong results, and a release check on typical test inputs ships it.
The arXiv Coding-Agent Research page documents the arXiv update.
39Updates
arXiv: Learning the Task or Learning the Verifier? Rethinking Criteria Exposure in Embodied Self-Improvement
Self-improving embodied agents are increasingly being developed and compared on robot simulation benchmarks.
The arXiv Coding-Agent Research page documents the arXiv update.
Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism
Reinforcement learning post-training improves the capabilities of large language models, but generating training responses is a major computational bottleneck.
The arXiv coding-agent research feed release records 2611.06984v1 for arXiv.
Adapting the Process of Adaptation: Darwinian, Lamarckian, and Agent-Directed Evolution in AI
Darwinian evolution relies on natural selection, Lamarckian evolution adds inherited learning, and agent-directed evolution allows agents to generate and evaluate behavioral variants and choose how extensively to search.
The arXiv coding-agent research feed release records 2611.07048v1 for arXiv.
Autonomous AI Agents Systematically Diverge From Human Preferences in Proof Formalization
Artificial intelligence (AI) systems have ushered in a new age of proof abundance, but helping humans understand these proofs requires new attention to the differences in how AI agents and humans pursue a proof-making task.
The arXiv coding-agent research feed release records 2611.06995v1 for arXiv.
arXiv: Scaling Neural Motor Decoding via Decoupled Behavioral Pretraining
Recent neural decoding approaches achieve strong performance by training on large collections of paired neural behavioral recordings, but such datasets are expensive, invasive, and difficult to scale.
The arXiv Coding-Agent Research page documents the arXiv update.
56Updates
New NVIDIA Cosmos 3 Open Models for Interactive World Simulation
Cosmos3-Nano-Sim-Auto: Stream synchronized camera and LiDAR observations for driving simulation.
The Hugging Face page documents the NVIDIA update.
The pypi registry records Google Gen AI SDK version 3.0.0.
UNVERIFIED — supplemental reports
UNVERIFIED Non-official information. Unverified report. The Information reports: California Sends AI Safety Lawmaker to Congress. The claim from The Information is not independently confirmed. Confirmation to watch for the The Information report: A first-party page from The Information that states it. The Information 2026-11-05T15:00:00Z
UNVERIFIED Non-official information. Unverified report. TechCrunch reports: Clutch, backed by Lightspeed and SignalFire, was last valued at $85 million, according to a source. The claim from TechCrunch is not independently confirmed. Confirmation to watch for the TechCrunch report: A first-party statement from the company involved that confirms it. TechCrunch 2026-11-06T06:28:32Z