OpenAI Codex
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 13
- GPT-6 Astra tops new physics benchmark, solving 49 of 50 problems
A team of 35 researchers from ICTS, Harvard, KU Leuven and Google DeepMind released TheoPhysEval, a benchmark of 50 unpublished theoretical physics problems with verifiable answers. OpenAI's GPT-6 Astra, running in Codex, solved 49 of 50 (98%) at under $3 per problem and about $116 for the full run; Anthropic's Claude Opus 5.5 in Claude Code scored 92%, Gemini 3.8 Flash 68%, and the open-weight DeepSeek-V4.1-Flash and GLM 5.3 reached 54% and 42%. The authors say web search made little difference and that agents still lack the "taste" to pick the right problems, so no comparable physics breakthrough has followed yet.
The Decoder · 🔥 4 - AI coding's real tragedy: software dies, but who gets to create?
A translated commentary (by Ian Bogost, originally in The Atlantic, republished by Huxiu) uses the claim that Adobe software was reverse-compiled by AI and released as open source to frame Silicon Valley's argument over whether software is now dead or worthless, with indie developer levelsio and others declaring the moat gone. Drawing on a year of building small apps like Beanslaw and Taptimer with AI, Bogost argues that AI coding does restore ordinary people's ability to write software for themselves, but that the most rewarding work comes from a back-and-forth partnership — you supply intent, the model implements, you judge and revise — rather than one-prompt delivery. He notes that agentic tools such as Claude Code and OpenAI Codex push toward full automation, while Apple's $99-a-year developer program and platform rules over things like haptics and distribution risk handing the creative process back to a few large companies; the Adobe episode is cited only as background to the debate, without verifiable detail.
虎嗅 AI · 🔥 0 - last30days agent skill adds native Codex plugin, new sources
The open-source AI agent skill last30days has updated to v3.11.1. According to the project, it has merged 175 pull requests since the v3.3 announcement in May — 122 of them from 52 community contributors across 15 releases. The skill lets an agent search Reddit, X, YouTube, Hacker News, Polymarket, GitHub and other platforms in parallel, rank results by upvotes, likes and money at stake, and synthesize one brief. This update adds a first-class OpenAI Codex plugin with renderer-aware citations, free keyless sources including arXiv, Techmeme and Digg, a rebuilt X search pipeline, and a doctor command that checks each source's health.
GitHub Trending · Python · 🔥 0 - Personal vs enterprise coding agents: eight functional gaps
A signed analysis published by TMTPost and written by Chen Chen, a Qi An Xin vice president, compares personal and enterprise coding agents across eight dimensions: task scope, context, permissions, execution sandboxing, toolchain integration, security and deployment, quality assurance, and collaboration and governance. The author argues the real difference is not feature count but constraints: personal agents aim to free individual capability, while enterprise agents must fold model uncertainty into a controllable, trustworthy and auditable framework — reading whole codebases, using tiered authorization, sandboxes and scoped credentials, private deployment and full audit logs. The article cites public material including Amplifying's State of the Coding Agent Market and a Gartner 2026 enterprise coding-agent Magic Quadrant, and says merged work from Copilot and Cursor is almost always reviewed while Codex and Devin have many PRs merged without a second reviewer; it states it is neutral analysis, not a product recommendation or investment advice.
钛媒体 · 🔥 0 - AI office agents: Tencent, ByteDance and Alibaba race for the next AI entry point
An analysis republished by Huxiu from Dolphin Research argues that with coding-scenario penetration near 70%, the AI entry-point battle is shifting from chatbots to task-completing office agents, noting that OpenAI's enterprise-side付费 has surpassed its consumer side at an annualized rate above $20 billion. It compares China's main contenders: Tencent's Workbuddy, which reported over 6 million daily active users in September; Alibaba's Qwen Office, built on DingTalk's 26 million-plus organizations and industry know-how; and ByteDance's Doubao Work, tied to Feishu and Volcano Engine with a field delivery team of roughly 60-70 people, while Zhipu's Zcode lacks a traffic入口. The author estimates a long-term global AI office TAM above $500 billion and about $45 billion for China, arguing that accumulated user-task trajectory data, rather than raw model performance, is the likelier durable moat.
虎嗅 AI · 🔥 0 - Coding-agent market splits into four quadrants, analysis argues
A neutral analysis published by TMTPost (authored by Chen Chen, a QAX vice president) divides the AI coding-agent market into four quadrants along individual vs. enterprise and China vs. international axes, arguing the four groups are almost different businesses in business model and competitive factors. It cites annualized revenue of over $2.5B for Claude Code, over $2B for Cursor, about $1B for Codex and about $5.25B... (specifically ~$525M) for Replit, and an Amplifying study of 166.7 million attributable GitHub outputs showing the top three tools take 93% combined, with Claude Code alone at 80%. The piece also cites Gartner's estimate of a $9.8B–$11B annualized enterprise AI coding-agent market, and concludes that individual tools compete on model capability, experience and cost while enterprise platforms compete on controllability, trust and auditability, with harness engineering the decisive factor behind "Agent = Model + Harness."
钛媒体 · 🔥 0 - REA tops 110K stars in a week as AI reverse-engineering wave stirs legal debate
REA (Reverse Engineer Anything) connects reverse-engineering tools such as Ghidra, IDA and JADX to coding agents like Claude Code and Codex through MCP, so a single prompt makes an agent inspect a binary or app package and return findings with evidence. Citing Star History data, the article says the project went from under 10,000 stars on October 6 to more than 110,000 on October 12 — roughly 100,000 stars in six days. It notes that open-source projects wiring AI into reverse-engineering tools, such as GhidraMCP and IDA Pro MCP, already existed, that several reverse-engineering experts call the hype overblown, and that sending decompiled output to model providers plus anti-decompilation clauses in user agreements can raise trade-secret, copyright and contract risks.
36氪 人工智能 · 🔥 0 - Composio's awesome-claude-skills hits GitHub trending with 1,000+ skills
Composio's awesome-claude-skills repository on GitHub collects more than 1,000 Claude Skills and plugins spanning document processing, development and code tools, data and analysis, business and marketing, writing, creative media, collaboration and security, and says they work across Claude.ai, Claude Code, Codex, Cursor, Gemini CLI and Windsurf. According to the repo, a Skill is a reusable instruction package containing a SKILL.md file with YAML frontmatter and Markdown instructions, loaded progressively: roughly 100 tokens per skill for name and description at session start, with the full body (typically under 5,000 tokens) loaded only when the agent judges it relevant. The list also promotes Composio's MCP Gateway, a single MCP endpoint for 1,000+ integrations with built-in authentication, team access controls and audit logs, plus a connect-apps plugin that lets Claude send emails, create issues and post to Slack; its README credits Anthropic with introducing the format in October 2025 and releasing it as an open standard in December 2025.
GitHub Trending · Python · 🔥 0 - Open-source Free Claude Code tops GitHub Trending with 61 providers
Free Claude Code (FCC) is an independent open-source project that says it is not affiliated with or endorsed by Anthropic, and lets users drive 11 coding agents — including Claude Code, Codex, Pi, OpenCode, Cline, Hermes and Aider — from a terminal, IDE, browser or phone. The project claims integrations with 61 “ToS-friendly” providers offering 1.3B+ free tokens per month, plus ordered fallback models, optional RTK/Headroom output compression and voice input via local Whisper or NVIDIA NIM. Free-tier availability and limits are controlled by each provider and may change.
GitHub Trending · Python · 🔥 0 - From Siri to Meta Muse: a 15-year history of AI assistants
This 36Kr-republished feature frames 15 years of AI assistants as five eras: Siri (2011) understanding spoken commands, Amazon's Echo/Alexa (2014) making computing ambient in the home, Microsoft 365 Copilot (2023) bringing LLMs into knowledge work, OpenAI's Codex CLI (April 2025) taking on whole engineering tasks, and Meta's personal agent Muse, unveiled at Connect in September 2026, which aims to act on a user's long-term intent. It says Meta negotiated a Manus deal in late 2025 worth more than $2 billion that did not close after roughly half a year of review, and that Muse is meant to remember context, run in the background and extend into WhatsApp, Meta's social apps and AI glasses. The author also describes China's 2026 office-AI race — Tencent WorkBuddy, Alibaba's Qwen Office and ByteDance's Doubao Work — as shifting from "embedding" AI to "taking over" tasks, and cites OpenAI figures that Codex could run roughly 25-hour tasks producing about 30,000 lines of code by late 2025.
36氪 人工智能 · 🔥 0 - Karpathy: a four-tier 'cognitive funnel' is splitting AI perception
Andrej Karpathy offered a "napkin math" estimate that roughly 75% of the world's population (about 6 billion people) has barely touched an LLM, while another 1–2 billion are occasional users of free ChatGPT-style products and therefore see AI as a "clumsy chatbot." He argues that some 20 million math and coding professionals are experiencing "agent vertigo" from paid frontier agents — especially OpenAI Codex and Claude Code — while at the very top only about 5,000 insiders have seen thousands of agents collaborate autonomously for weeks, hunt zero-days and push frontier mathematics. He attributes the split to three factors: the ambition and scale of your problems, the verifiable rewards available in code and math that suit reinforcement learning, and the access gap between free, paid and internal tiers. The population figures are his own estimates, not audited statistics.
36氪 人工智能 · 🔥 0 - Agency Agents ships a native desktop app for its agent roster
Agency Agents is a growing collection of AI agent personas that started from a Reddit thread and months of iteration, spanning engineering categories such as frontend, backend, DevOps, security and data, with each agent given its own identity, workflows and deliverables. The project now offers a native macOS, Linux and Windows app that browses the full roster and installs agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen and Osaurus in one click, alongside script-based installs and auto-updates. The repo describes the agents as "production-ready," a claim that comes from the project itself rather than independent testing.
GitHub Trending（每日） · 🔥 0 - Why AI Applications Struggle to Make Money
This opinion article argues that standalone AI applications will struggle to sustain profits because single-point capabilities such as coding, translation and presentation generation have low barriers and switching costs. It says the stronger business model is enterprise-focused workflow capture: using AI to take over complete business processes, tie value to measurable P&L outcomes, and embed AI into products, services and operations rather than selling consumer subscriptions.
虎嗅 AI · 🔥 0
Experience and discussion from the community
Share my OpenAI Codex experienceAsk about OpenAI Codex
Nobody has shared their experience with OpenAI Codex yet.