Skip to main content
Back

OpenAI releases GPT-5-Codex with 7-hour agentic coding

Anthropic brings Claude to Xcode, OpenAI updates ChatGPT tone, ElevenLabs launches services, Google tests agent economie

September 16, 2025 4 min read

Hey

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers.


Every day, we identify and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.


Here’s today’s roundup:

SUMMARY

Read time: 4 min 20 sec

Top News

▪ OpenAI debuts GPT-5 Codex for agentic coding

Trending Signals

▪ Anthropic brings Claude Sonnet 4 into Apple’s Xcode 26

▪ OpenAI lets users change ChatGPT tone without affecting abilities

▪ ElevenLabs offers managed content services with fast, competitive pricing

▪ Anthropic shares open dataset on geographic, occupational, and enterprise AI use

▪ Google paper shows AI agents can trade and collaborate in virtual economies

Top Reads

▪ Simon Willison: Models now write their own prompts

▪ Vercel explains why MCP tools should target LLMs, not developers

▪ Subagents cut context pollution in multi-tool agents

How To

▪ Automate code reviews in GitHub with OpenAI's Codex

TOP_NEWS

OpenAI launches GPT-5 Codex with agentic coding and code review features, can run 7 hours independently at a time

18,385 Likes

alpha_signal_image_2

OpenAI has upgraded Codex with GPT-5-Codex, a GPT-5 variant optimized for real-world software engineering. Codex is now faster, more reliable, and better at independent coding across IDEs, terminals, web, GitHub, and mobile.


Key Features

  • Agentic coding: Handles quick pair-programming or runs independently for 7+ hours on large refactors.

  • Code review: Navigates repos, runs code/tests, and flags critical bugs with higher precision.

  • Dynamic reasoning: Adapts time spent “thinking” based on task complexity.

  • Tool use: Supports images, web search, MCP integration, and browser-based inspection.

  • Everywhere: Works in CLI, IDEs (VS Code, Cursor), GitHub, ChatGPT web, and mobile.

Updates Across Codex

  • CLI: Revamped for agentic workflows; supports screenshots, diagrams, tool calls, and progress tracking.

  • IDE extension: Integrated into VS Code and Cursor; keeps context synced across cloud/local.

  • Cloud: Faster completions (90% faster via caching), auto environment setup, visual debugging with screenshots.

Technical Improvements

  • Training: Tuned on large-scale, real-world engineering tasks (refactors, full projects, debugging).

  • Steerability: Follows AGENTS.md instructions closely, producing higher-quality code with minimal prompting.

  • Improved performance, e.g., scoring 72.8% on SWE-bench, close to GPT-5’s 74.5%.

  • Efficiency: Uses 93.7% fewer tokens on simple tasks, but spends 2× more reasoning time on complex ones.

Availability

  • Plans: Included with ChatGPT Plus, Pro, Business, Edu, and Enterprise.

  • Scaling: Usage scales by plan; Business/Enterprise can purchase credits or share pools.

  • API: GPT-5-Codex will be available soon for Codex CLI via API key.

Try now

TOP_SIGNALS

Anthropic’s Claude Sonnet 4 now fully integrated into Apple’s Xcode 26 coding environment

1,649 Likes

OpenAI introduces customizable ChatGPT personalities for Plus, Pro, Team

5,248 Likes

ElevenLabs launches Productions, offering human-edited dubbing, captions, transcripts, audiobooks

832 Likes

Anthropic reports API users automate 77% of tasks, releases open dataset on geographic and enterprise AI adoption patterns

1,783 Likes

Google DeepMind proposes sandbox economies where autonomous AI agents negotiate, trade, and coordinate resources globally

2,448 Likes

TOP_READS

Simon Willison: Models now write their own prompts

1,358 Likes

The article by Simon Wilson notes that models are getting better at creating prompts. It attributes this to large prompt-rich datasets, pre-training exposure to high-quality prompts, and practices like Claude Code’s tool use.

Vercel explains why MCP tools should target LLMs, not developers

529 Likes

Vercel argues the second wave of MCP design should prioritize LLMs. By bundling multi-step operations into single tools and handling domain logic internally, agents become more robust and context-aware.

Subagents cut context pollution in multi-tool agents

642 Likes

Subagents work under an orchestrator which breaks down user requests, assigns tasks to subagents with limited tools and isolated context, then aggregates results. Explicit ones offer stability; on-the-fly ones offer flexibility.

HOW_TO

Automate code reviews in GitHub with OpenAI's Codex

alpha_signal_image_3

This guide shows how to run code reviews in GitHub using Codex cloud.


Step 1 - Set Up

Connect your GitHub account to Codex cloud. In Codex settings, enable Code review for your repository.


Step 2 - Trigger Review in PR

Comment @codex review on a pull request. Codex posts review comments automatically.


Step 3 - Run Other Tasks

Use @codex <instruction> in PR comments to request tests, refactors, or other tasks.

More info

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.


Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 250,000+ AI developers?

Work with us