With OpenAI’s GPT-5-Codex set firmly in the limelight, Anthropic looks to try snatch it away with the release of Claude Sonnet 4.5, its next frontier AI coding model.
Our other key takeaways include:
DeepSeek releases an “intermediate” V3 model en route to its next generation
OpenAI’s parental controls go live on ChatGPT for a safer under-18 experience
Anthropic announces plans to 3x its workforce and open new offices in Japan and Europe
Join us at AI Tangle as we untangle this week's happenings in AI!
THE BIG AI STORY
Credits: Anthropic
Late into the Monday workday, Anthropic launched its latest frontier AI coding model called Claude Sonnet 4.5, claiming new SOTA-level real-world performance. According to Anthropic, Sonnet 4.5 comes with a leap in reliability: where other models could create prototype applications, Sonnet 4.5 can create entire production-ready ones. How does Sonnet 4.5 fare against the others at the frontier, and what sets it apart from the rest?
What makes it special?
When stacked up against the competition, Sonnet 4.5 improves upon its predecessors and remains best of the rest across coding benchmarks, such as SWE-bench Verified and Terminal-bench. Sonnet 4.5's autonomous coding length, however, is what Anthropic claims makes its model unique, reporting observations of it "maintaining focus for more than 30 hours on complex, multi-step tasks." Together with performance and coding length improvements, Sonnet 4.5 also debuts with lower sycophancy and deception rates, as well as improved resistance to prompt injection.
The model is available via the Claude API and in the Claude AI webapp, available for the same price as Sonnet 4.0 for developers.
5 QUICK HITS
Ahead of China's National Day holiday, AI wonderchild DeepSeek has released a new, "experimental" version of its V3 foundation model, the aptly-named DeepSeek V3.2-Exp. The model massively improves on transformer architecture efficiency and long-context input performance, while at the same time reducing compute costs thanks to a mechanism it calls wDeepSeek Sparse Attention. After announcing "50%+" cuts to API costs, DeepSeek's V3.2-Exp model is available on Hugging Face, ModelScope, and the startup's own website and app.
OpenAI is introducing parental controls for ChatGPT following a not-so-distant incident where a California family sued the chatbot maker for influencing their teenage son's suicide. The new measures allow parents to link accounts, limit exposure to sensitive content, and decide whether ChatGPT remembers past conversations. It also enables parents to set "quiet hours," disable ChatGPT's voice mode, and even receive alerts if the system detects "signs of acute distress." Although OpenAI stressed that the system is not perfect and can issue false positives, they are nonetheless a step forward for broader online safety.
According to a recent blog post, Anthropic announced plans to triple its global workforce and expand its applied AI team fivefold this year as demand for Claude outside the US surges, accounting for 80% of its usage. The company's customer count has risen from a mere 1,000 to 300,000 in two years, fueling revenue growth from $1 billion to $5 billion this year alone. It also plans to open its first Asian office in Tokyo, along with new ones across Europe, spearheaded by the recently-appointed Managing Director of International, Chris Ciauri.
Meanwhile, in contrast to Anthropic, Germany's Lufthansa Group stated on Monday that it will cut 4,000 administrative jobs by 2030 in favor of streamlining its work with AI, digitalization, and closer integration of member airlines. The airline group claims to expect "significantly increased profitability" due to increase in demand and a limited flight supply, despite all the cuts. Lufthansa is also planning its largest-ever fleet renewal by adding over 230 new planes, including 100 long-haul ones, by the end of the decade.
Microsoft looks to bring "vibe working" to its Office suite by introducing what it calls Agent Mode in Excel and Word. Powered by GPT-5, Agent Mode breaks complex tasks into agentic step-by-steps and can be used for generating spreadsheets, documents, monthly reports, etc., with full auditability. Meanwhile, a companion Office Agent in Copilot chat, leveraging Anthropic's models, can create entire PowerPoint decks and Word documents from just prompts. The initial rollout is web-only through Frontier for Microsoft 365 Copilot and Personal/Family subscribers, with desktop support expected later.
4 AI TOOLS
Fellow - Built for hybrid and remote teams, Fellow is an AI meeting management solution for those who want to move faster and smarter.
Latitude - Level up your prompt game with Latitude, an open-source engineering platform that evaluates your prompts to chatbots and helps you refine them.
Scottie - Scottie turns plain prompts into working, cloud-run AI agents leveraging top industry models in 5 minutes that can easily integrate into over 200 productivity tools, from Slack to Google Sheets.
Study Space - Transform how you study and access knowledge with Study Space, a platform to turn documents and links into fully fleshed out and immersive courses to take at your own pace.
AI EXTRA READ
Failing to Understand The Exponential, Again (4-min read)
Is AI another dotcom-like bubble, or will it simply never live up to the hype? Former Google AI researcher and current Anthropic one Julian Schrittwieser believes neither, claiming that many fail to understand the significance of AI's current capabilities, what it started out as in the past, and its future progress.
MORE FROM THE AIE NETWORK
📻 AI Confidential Podcast - Are LLMs Dead?
💼 The Artificially Intelligent Enterprise - The True Cost of AI
💡 AI CIO - The Paradox of AI Reports
🎯 The AI Marketing Advantage - Your New Morning Habit: ChatGPT Pulse
What did you think of this newsletter? Let us know!
Your AI Sherpa,
Mark R. Hinkle
Publisher, The Artificially Intelligent
Enterprise (TheAIE) Network
Connect with me on LinkedIn
Follow me on X
