Claude Opus 4.7 Widens Gap

Codex Now Controls Your Mac

Welcome back!

Anthropic just released Claude Opus 4.7, widening the gap in agentic coding. The model now leads coding benchmarks, reads screenshots more accurately, and follows instructions more reliably. But there’s a twist: it consumes 30% more tokens, raising costs. Let’s unpack…

Today’s Summary:

  • 🚀 Anthropic launches Claude Opus 4.7

  • 🖥️ Codex now controls your Mac

  • 🔥 Claude Design enters creative stack

  • 🏆 Qwen3.6 beats Gemma 4

  • 🧬 GPT-Rosalind for life sciences

  • 🔎 Chrome AI Mode

  • 🛠️ 2 new tools

TOP STORY

Anthropic releases Claude Opus 4.7

The Summary: Anthropic launched Claude Opus 4.7 with big improvements in autonomous software engineering while intentionally reducing cybersecurity capabilities. The model scores 64.3% on SWE-bench Pro (up from 53.4% for Opus 4.6), processes images at triple the resolution, and follows instructions more reliably. However, a new tokenizer increases token consumption by 30%, raising actual session costs.

Key details:

  • #1 in Code Arena WebDev

  • #1 in GDPval-AA benchmark for performance on real-world tasks

  • The token consumption change may lead to users burning through limits faster or paying more per session

  • Image input resolution jumped from ~1.2 to 3.75 megapixels (2,576 pixels on long edge), enabling better screenshot reading

  • New xhigh effort level introduced between "high" and "max", with Claude Code defaulting to xhigh

Why it matters: Opus 4.7 holds the top spot with a 130-point lead over GPT-5.4 and Gemini 3.1 Pro in Code Arena and a 60% win rate on real-world work tasks. The gap matters because agentic coding is where enterprises currently deploy these models at scale.

FROM OUR PARTNERS

The Builder Conference

Built for builders. Not buzzwords. San José 2026

500+ speakers. 18 content tracks. Workshops, masterclasses, and the people actually shipping the tools you use every day. WeAreDevelopers World Congress — September 23–25. Use code GITPUSH26 for 10% off.

OPENAI

OpenAI updates Codex to control your Mac in the background

The Summary: OpenAI released a major Codex update that lets the AI coding tool take control of your Mac while you work on other tasks. The desktop app now runs multiple agents in parallel, uses an in-app browser, generates images, and remembers context across sessions. With 3 million weekly users, Codex expands beyond terminal coding into a workspace that handles the entire software development lifecycle.

Key details:

  • Works by seeing, clicking, and typing

  • Over 90 new plugins connect Codex to JIRA, GitLab, Microsoft Suite, Slack, Gmail, and Notion for cross-platform task management

  • New Heartbeat Automations schedule future work and wake up automatically to continue tasks across days or weeks

  • Memory feature retains preferences from previous sessions

  • OpenAI acquired the Sky team (former Apple employees) to build the macOS cursor technology

  • Granting agents full computer access remains a severe security problem

  • Computer use initially available on macOS, excluding EU/UK

Why it matters: OpenAI frames this as building toward a "Super App" but the real story is agents getting persistent memory and the ability to operate a browser. Codex can now scan Slack, Gmail, and Notion to build your daily agenda, competing with traditional software. Traditional SaaS will need agent-native versions optimized for agents. This creates an opportunity for companies that adapt first to capture the agent-mediated market.

FROM OUR PARTNERS

A Smarter Way to Read the News

Tired of news that feels like noise?

Every day, 4.5 million readers turn to 1440 for their factual news fix. We sift through 100+ sources to bring you a complete summary of politics, global events, business, and culture — all in a brief 5-minute email. No spin. No slant. Just clarity.

ANTHROPIC

Anthropic launches Claude Design

The Summary: Anthropic launched Claude Design, an AI tool that creates designs, prototypes, and presentations through conversation. The product runs on Claude Opus 4.7 and targets both designers seeking rapid exploration and non-designers who need to visualize ideas. Users describe what they want, and Claude generates editable outputs in PPTX, PDF, Canva, or hands them off directly to Claude Code for implementation.

Key details:

  • Powered by Claude Opus 4.7, which handles images up to 2,576 pixels

  • Can read codebases to automatically apply brand design systems

  • Brilliant's senior product designer reported complex pages took 20+ prompts in competing tools but only 2 prompts in Claude Design

  • Mike Krieger (Anthropic's CPO) resigned from Figma's board 3 days ago

  • Available now for Pro, Max, Team, and Enterprise subscribers at no additional cost within existing limits

Why it matters: Anthropic just declared war on the design tool stack. The company now has a complete creative loop with Claude Design for mockups, Claude Code for implementation, and Cowork for project management. Anthropic is building an application empire on top of its models, and the recent $800 billion valuation offers reflect investor belief that Anthropic is headed toward the full application stack beyond the AI model itself.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/