AI 速递 2026-07-19
生成时间：2026/7/19 09:24:44（UTC: 2026-07-19T01:24:44.880Z）
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget
Section titled “LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget”👍 172 · arXiv
A growing gap separates inference context lengths from RL post-training: inference systems are approaching million-token contexts, while post-training workloads often remain at 256K tokens or below an…
Boogu-Image-0.1: Boosting Open-Source Unified Multimodal Understanding and Generation
Section titled “Boogu-Image-0.1: Boosting Open-Source Unified Multimodal Understanding and Generation”👍 125 · arXiv
We introduce Boogu-Image-0.1, an open-source unified multimodal understanding and generation model family, comprising Base, Turbo, Edit, and Edit-Turbo variants. It delivers competitive performance in…
VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding
Section titled “VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding”👍 120 · arXiv
Recent advances in video understanding have spanned motion, long video, and streaming interaction, driving this field toward real-world applications. Despite this progress, current open-source models …
Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning
Section titled “Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning”👍 90 · arXiv
Reinforcement learning with verifiable rewards without human-annotated data, often referred to as zero RL, has emerged as a powerful paradigm for eliciting chain-of-thought reasoning. However, due to …
SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning
Section titled “SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning”👍 83 · arXiv
Large language models are increasingly trained as interactive agents for long-horizon tasks involving multi-turn interaction, tool use, and environment feedback. Outcome-based reinforcement learning (…
OpenClaw v2026.7.2-beta.3
Section titled “OpenClaw v2026.7.2-beta.3”2026.7.2
Section titled “2026.7.2”Highlights
Section titled “Highlights”- Remote coding sessions: run Control UI sessions on cloud workers, open Codex and Claude catalog sessions in terminals on their owning hosts, and resume OpenCode and …
链接：https://github.com/openclaw/openclaw/releases/tag/v2026.7.2-beta.3
CrewAI 1.15.4
Section titled “CrewAI 1.15.4”What’s Changed
Section titled “What’s Changed”Features
Section titled “Features”- Promote Skills Repository out of experimental status
Documentation
Section titled “Documentation”- Add Flows in Studio documentation
Contributors
Section titled “Contributors”@jessemiller, @joaomdmoura, @vinibrsl…
链接：https://github.com/crewAIInc/crewAI/releases/tag/1.15.4
OpenAI Codex CLI rust-v0.144.6
Section titled “OpenAI Codex CLI rust-v0.144.6”Bug Fixes
Section titled “Bug Fixes”- Refreshed bundled instructions for GPT-5.6 Sol, Terra, and Luna, and corrected their context windows to 272,000 tokens. (#33972, #34009)
Changelog
Section titled “Changelog”Full Changelog: https://github.c…
链接：https://github.com/openai/codex/releases/tag/rust-v0.144.6
Kimi: Threat or menace?
Section titled “Kimi: Threat or menace?”Chinese company Moonshot AI released a new version of its Kimi model this week, prompting concern about “full AI communism.”
Neil Rimer thinks the AI money is coming back out
Section titled “Neil Rimer thinks the AI money is coming back out”Neil Rimer, the venture capitalist who co-founded Index Ventures, predicts the historic wealth AI is generating in Silicon Valley will have to be redistributed, voluntarily or involuntarily.
Vertu wants executives to pay $6,880 for an AI agent — here’s how it actually performs
Section titled “Vertu wants executives to pay $6,880 for an AI agent — here’s how it actually performs”From AI workflows to battery life and security, here’s what it’s really like to live with Vertu’s luxury foldable every day.
Databricks hits $188B valuation, extending its run as AI’s favorite second act
Section titled “Databricks hits $188B valuation, extending its run as AI’s favorite second act”Databricks has remade its image into an AI company and has published research on the cost savings of open weight AI models for coding.
The Zoom hack that says, ‘Don’t record me’
Section titled “The Zoom hack that says, ‘Don’t record me’”If every meeting, watercooler conversation, and date gets transcribed and summarized, who’s actually reading any of it?
Agility Robotics plants its flag in Tesla’s backyard
Section titled “Agility Robotics plants its flag in Tesla’s backyard”Agility is opening a new training center for its Digit robots in Fremont, California.
AI-driven memory crunch jolts India’s smartphone market
Section titled “AI-driven memory crunch jolts India’s smartphone market”India’s smartphone slowdown highlights how the AI boom is reshaping consumer electronics, from pricing and demand to corporate strategy.
How Apple’s big lawsuit could disrupt OpenAI’s IPO plans
Section titled “How Apple’s big lawsuit could disrupt OpenAI’s IPO plans”Apple filed a trade secrets lawsuit against OpenAI last Friday, and it’s not messing around. The complaint alleges a pattern of misconduct reaching all the way up to OpenAI’s chief hardware officer and claims more than 400 former Apple employees now work at the c