20 November 2026
86 new posts across 9 AI lab and practitioner sources — 8 worth your time.
GPT‑6.2 Astra and GPT‑6.2 Sol
OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol in the API. GPT-6.2 Astra is priced at $8 per million input tokens and $40 per million output tokens, representing an 80% price point relative to GPT-6 Astra.
Simon Willison · 20 November 2026 · Read the original →
TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient
Microsoft Research released TokenEconomizer, an open-source plugin for coding agents. It records repo-specific context and tool execution outcomes across developer sessions to avoid redundant code indexing and failed steps in subsequent runs.
Microsoft Research · 20 November 2026 · Read the original →
NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance
NVIDIA released MLPerf Training v6.1 benchmark results introducing the Vera Rubin architecture. In its initial MLPerf Endpoints entry, the platform measured over 10x higher throughput per watt for agentic AI workloads.
NVIDIA · 20 November 2026 · Read the original →
Understanding Positional Embeddings in Modern LLMs
This technical guide details how transformer architectures process token order, focusing on Rotary Position Embeddings (RoPE). It includes visual diagrams, mathematical derivations, and a complete PyTorch code implementation.
Ahead of AI · 20 November 2026 · Read the original →
External sandboxes for coding agents
Simon Willison and Prime Radiant detailed isolation patterns for local LLM code execution agents. The write-up addresses defensive boundaries required to run local models without risking file system destruction or prompt injection attacks on the host machine.
Simon Willison · 20 November 2026 · Read the original →
Inside the Codex sandbox
OpenAI detailed the isolation architecture running behind Codex on developer machines. The post explains how the engine enforces filesystem permissions and network boundary controls across macOS, Linux, and Windows environments.
OpenAI · 20 November 2026 · Read the original →
Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics
Cloud provider Lambda integrated NVIDIA Photonics co-packaged optics (CPO) switches into its GPU infrastructure. The optical setup connects NVIDIA GB300 NVL72 racks across clusters exceeding 10,000 GPUs to reduce power usage and extend mean time between network failures.
NVIDIA · 20 November 2026 · Read the original →
Anthropic IPO Prices at $83 Per Share
Anthropic priced its initial public offering at $83.00 per share on November 18, 2026, with shares expected to begin trading on Nasdaq on November 19, 2026, under the symbol "ANT". The offering is expected to close on November 20, 2026.
Anthropic · 20 November 2026 · Read the original →
We are spending a lot of engineering just keeping the code-writing agents contained.
8 stories, every Friday
Published here every week. Follow by RSS to get it as it lands.