AI · Daily

Stellar DailyVOL. 2026.07.25 · 8 STORIES

Saturday, July 25, 2026

Complete edition, collected through 24:00

Today's main signal is the release of Claude Opus 5, which approaches Fable 5-level intelligence with dramatically improved long-running agent and coding capabilities, plus a 50% price reduction. Hailed as the most prompt-injection-resistant model to date, its ecosystem tool Claude Code immediately defaults to Opus 5 and introduces sub-agent nesting. Meanwhile, AI personality emerges as a new competitive edge (Cognition acquires Poke), while OpenAI's AI keyboard sparks mixed reactions, highlighting the gap between utility and user understanding.

01

Models

3
  1. Claude Opus 5 Released: Major Improvements in Long-Running Agents and Coding
    T1Anthropic:News(网页)

    Anthropic has officially launched Claude Opus 5, a step-change upgrade to the Opus series optimized for long-running agent tasks, with significant improvements in code generation, understanding, and professional work processing, aiming to empower more complex automation workflows and enterprise-grade applications.

  2. Anthropic Opus 5: The Most Prompt-Injection-Resistant Model Yet
    T2Simon Willison 博客(Atom)

    According to Boris Cherny, Anthropic's latest Opus 5 model has achieved a significant breakthrough in prompt injection defense, making it the least injection-prone model from the team so far. Although specific evaluation scores have not been released, this progress is crucial for developers building secure applications with large language models, marking a key step forward in AI safety.

  3. Claude Opus 5 Released: Near Fable 5 Intelligence at Half Price, Proactively Builds Computer Vision Pipeline
    T2Simon Willison 博客(Atom)

    Anthropic has released Claude Opus 5, claiming it approaches the intelligence of flagship model Claude Fable 5 at half the price and surpasses Fable 5 on the Artificial Analysis leaderboard. It is priced the same as Opus 4.8 and offers a fast mode. A standout anecdote: given a drawing without direct access, the model wrote its own computer vision pipeline to extract geometry from pixels and reconstruct a 3D model. In cybersecurity, Opus 5 nears Mythos 5 in vulnerability detection but is untrained in exploitation. Anthropic also published a prompting guide.

02

Products

4
  1. Claude Code v2.1.219 Released: Defaults to Opus 5 and Introduces Nested Subagents
    T1.5Claude Code:GitHub Releases(RSS)

    Claude Code v2.1.219 updates the default Opus model to Claude Opus 5 with 1M context and fast mode pricing at $10/$50 per million tokens. It adds a strict network allowlist for sandboxed commands, increases default subagent nesting depth to 3, introduces workflow size guidelines, and fixes numerous bugs including improved MCP error reporting. This release enhances developer productivity and security.

  2. Corv v1.1 Released: SSH Execution Client Built for AI Agents
    T2Hacker News 热门(Algolia · 中文化)

    Corv Client is an SSH client designed for both AI agents and humans, addressing the challenge of remote command execution for AI. Version 1.1 brings improved connection stability, enhanced authentication support, and better agent integration, enabling AI to securely and efficiently manage remote infrastructure via SSH. The project is open source on GitHub, suitable for developers, DevOps engineers, and teams building automated agents.

  3. Cognition Acquires Poke: AI Personality Becomes a Competitive Advantage
    T2TechCrunch:AI(RSS)

    AI coding startup Cognition has acquired Poke, an AI assistant known for texting like a friend, in a deal valued in the low nine figures. The acquisition integrates Poke's conversational style into Cognition's coding agent Devin, signaling that interaction design is becoming as crucial as underlying AI models for competitive advantage.

  4. Hands-on with OpenAI's New AI Keypad: Fun for Coders, Puzzling for Everyone Else
    T2TechCrunch:AI(RSS)

    OpenAI has unveiled a novel AI keypad designed to give coders quick access to AI-assisted features like code completion and function generation. After hands-on testing, the device feels intuitive and powerful for developers, but its purpose and utility remain mystifying for non-technical users. The keypad integrates OpenAI's latest models, supports voice and key triggers, and boasts a compact design.

03

Industry

1
  1. Prentis, new AI lab co-founded by Reid Hoffman and Marc Pincus, in talks to raise $100M
    T2TechCrunch:AI(RSS)

    Prentis, a new AI lab co-founded by LinkedIn co-founder Reid Hoffman and Zynga founder Marc Pincus, is in talks to raise $100 million. The lab bets that automating routine computer tasks will soon outpace coding as AI's biggest use case. It aims to develop AI systems that understand and execute complex workflows, though specific technical details have not been disclosed yet.

04

Research

0
05

Tips & Opinions

0

Flashes

27
Events today
1
First-hand reports
5
New models
6
Sources

Archive

July 2026