OpenAI Releases Images 2.6 With SVG Export, GIFs, and Faster Generation
OpenAI released Images 2.6 on November 4, with access for ChatGPT and Codex subscribers across all tiers and an API for developers. The report describes SVG export and GIF generation, plus two API versions: Flare, oriented toward speed, and Sunburst, which OpenAI says better preserves details across edits at medium or high quality. It also gives API pricing and notes that the performance claims have no third-party test results yet.
Model releases and updates
Kimi · X
Kimi Releases K3.1 Pro Model Weights
Kimi announces the release of Kimi K3.1 Pro's model weights and links to the model repository on Hugging Face. The post adds, “We'll keep advancing open intelligence.”
Dharma Field Extractor Turns Document Images into Structured JSON
Dharma-AI introduces the Dharma Field Extractor, a 2-billion-parameter model that converts document images into structured JSON, either by extracting fields it identifies or by filling a user-specified schema. The authors report a score of 0.834 across 1,638 benchmark cases, above four tested frontier configurations, with self-hosted GPU costs 6 to 147 times lower than the compared APIs.
Product releases and updates
AWS Machine Learning BlogFirst-party
Amazon Quick Smart Approvals Adds Human Review for Consequential Actions
Amazon Quick’s Smart Approvals reviews eligible agent actions against the user’s request and conversation context, approving some automatically and returning others for human review. Existing permission controls and mandatory approvals remain in force, and each automatic approval applies only to the specific action and parameters reviewed. The article also describes audit logs and a pilot approach for evaluating the feature.
vLLM · X
vLLM Brings Day-Zero Serving for Kimi K3.1 Pro
vLLM says it supports serving Kimi K3.1 Pro on day zero. The post cites a 1M-token context window for long coding sessions, 3D creation, and research, and reports a 65.7% score on Terminal-Bench 4.0 with the Claude Code harness. It also says vLLM provides day-zero serving across GPUs and TPUs, plus RL support.
vLLM · X
vLLM Releases Version 0.33.0 With New Models and Runtime Updates
vLLM released version 0.33.0, citing 927 commits, 389 contributors, and 122 first-time contributors. The update adds support for Reflection Beam-1.0, Step-5 Preview, Nemotron Parse 3.5, and MiniCPM-OCR 1.0, alongside PyTorch 2.14 wheels that support CUDA 13.0 only. The post also lists speculative decoding changes, NVIDIA decode updates, a new AWS EFA image, and ROCm 10.0 as the default.
Industry
Ars Technica · AI
US Election Results Put AI and Data Center Expansion Under Pressure
Voters backed candidates and measures favoring tighter restrictions on AI and data centers in the US, including 38 of 44 ballot measures tracked across 10 states. The article reports projections that delays could affect future projects, particularly in Texas, while tech industry representatives argued the results did not show a voter mandate to crack down on AI. Other analysts cited polls showing opposition to data centers and voter support for AI rules.
OpenAI Pipeline Finds and Patches 948 Linux Kernel Privilege Escalation Vulnerabilities
OpenAI says it built an autonomous pipeline that found, validated, and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel.
OpenAI to Acquire Pierre Computer Company, Inc.
OpenAI says it will acquire Pierre Computer Company, Inc., expanding Codex’s tools for building, understanding, and reviewing code.
Research
Anthropic · XAnthropic Describes Using Claude to Map Bacterial Vault Structures
Anthropic says its biology research team used Claude to map the first 3D structure of a vault particle from bacterial life. The team found that similar proteins from bacteria and the viruses that infect them can assemble into vaults, which may make their function easier to investigate. Anthropic says some stages took days rather than the weeks to months manual work would have required, and describes mistakes addressed with machine learning researchers.
Tips and perspectives
Nathan Lambert · XRicursive AI Releases Ripple-65B Preview, a Dense Model Trained From Scratch
Ricursive AI says it has released Ripple-65B Preview, a 65B dense LLM pretrained from scratch. The company claims Ripple sets a state of the art for US open-weight models on agentic coding tasks, surpassing several larger models. Nathan Lambert praises the project and connects it to the OLMo lineage.
AWS Machine Learning BlogFirst-party
How to Connect Decision Models to Amazon Bedrock AgentCore Evaluations
AWS explains how to connect TypeSafe’s hosted Jev and the open-source Strands Decider 2B to Amazon Bedrock AgentCore Evaluations through code-based evaluators. The walkthrough builds an HR Assistant example and shows how to map model probabilities into evaluation results, then validate the checks against human labels and built-in evaluators on the same traffic. The post recommends using decision models for narrow checks alongside broader LLM evaluations.
Tomorrow at 08:00, a briefing like this arrives in your inbox.
The same kind of briefing as the one above: a few stories worth reading. Free. Mail starts after you confirm, and every letter has a one-click unsubscribe. Read a past issue · 2026-11-04