- One more Thing in AI
- Posts
- Siri is going to get an Upgrade and Kimi model beats top models
Siri is going to get an Upgrade and Kimi model beats top models
Open AI crosses 1 M Business customers
Renjit Philip
November 10, 2025 • Estimated Reading Time: 11 minutes
Date: 2-Nov-2025
Hey AI enthusiast,
And just when you think you’ve seen it all…there’s always One More Thing in AI.
Table of Contents
Also: introducing you to our podcast edition for those of you who want to listen to it on the go. Check it out on your favorite platforms:
I hope you enjoy this edition!
Best,
Renjit
PS: If you want to check out how to implement AI agents in your business and get more revenue with the same number of employees, speak to me:
OpenAI Reaches 1 Million Business Customers and Redefines How Work Gets Done
ChatGPT started as a personal tool. Now it’s becoming the backbone of how companies work.
Summary:
OpenAI has passed 1 million business customers, making it the fastest-growing business platform ever. These include companies paying for ChatGPT for Work and others using OpenAI models through its developer tools. My take is that they had to come out with this information after the news came out about Anthropic’s business usage numbers, which were pretty high compared to OpenAI.
With over 800 million weekly ChatGPT users, companies can roll out AI faster because their employees already know how to use it. ChatGPT for Work now has 7 million seats, a 40% jump in two months, while ChatGPT Enterprise grew 9x in a year.
What’s new for businesses:
🧠 Company Knowledge: A GPT-5 model can now search across Slack, SharePoint, Google Drive, and GitHub, giving clear answers and citing sources.
⚙️ Codex Growth: Code generation and automation use is up 10x since August. Cisco says it cut code review time by 50%.
🤖 AgentKit: Lets teams build and deploy AI agents quickly. Carlyle used it to speed up due-diligence work by half and improve accuracy by 30%.
🎨 Multimodal Tools: New APIs for text, image, video, and voice mean one system can now handle every format of work.
Proof that AI delivers results:
A Wharton study found 75% of enterprises see positive ROI, while fewer than 5% report losses. A few use cases:
• Indeed’s AI-driven “Invite to Apply” raised job applications by 20%.
• Lowe’s gave store employees AI guidance across 1,700 stores.
• Intercom’s “Fin” agent cut development time from months to days.
• Databricks now brings OpenAI intelligence directly to company data.
What’s next:
More businesses are starting to build with OpenAI, not just use it.
• Canva, Figma, Zillow, and Spotify have linked their apps to ChatGPT.
• Shopify, Walmart, PayPal, and Salesforce are using the Agentic Commerce Protocol (ACP) to bring conversational shopping into ChatGPT. I wrote about Agentic Commerce Protocol in a previous edition.
OpenAI isn’t just supplying tools anymore. It’s building the new operating system for business.
🌍Apple Bets on Google’s Gemini to Finally Fix Siri
For years, Siri has lagged behind Alexa, ChatGPT, and even Google Assistant. Now, Apple is taking an unexpected path to catch up by licensing Google’s Gemini model.
Bloomberg reports that Apple has finalized plans to use a custom 1.2 trillion-parameter version of Gemini for Siri’s long-awaited overhaul. The deal, worth roughly $1 billion per year, signals Apple’s quiet but serious move into the AI race.'
Source: Apple website
Here’s what’s changing:
• Gemini will power Siri’s brains for tasks like summarization and multi-step planning.
• The model will run on Apple’s Private Cloud Compute, ensuring personal data stays private. This is important for privacy focused folks like me!
• Apple tested models from OpenAI and Anthropic before settling on Google’s.
• For context, this new Gemini version dwarfs Apple’s in-house 150B parameter Apple Intelligence model.
Interestingly, the partnership won’t be branded publicly. Google will remain “behind the scenes,” while Apple presents the new Siri as its own upgrade.
The rollout could start next spring, giving Siri a long-overdue intelligence boost.
But here’s the twist:
Apple sees Gemini as a temporary fix- a bridge while it builds its own model. Given years of delays, internal turnover, and lack of clear AI leadership, that’s easier said than done.
For now, though, Siri might finally stop misunderstanding what users want. And it’s Google, not Apple, making that happen.
The next time Siri sounds smarter, it’s likely thanks to Mountain View, not Cupertino.
When the Market’s Hottest Stocks Are Getting a Cold Bet
Think the AI boom will just keep soaring? One investor says maybe not.
• 📌 The investor is Michael Burry, known for calling the 2008 sub-prime crash.
• He used put options (bets on a stock’s decline) on Nvidia ($NVDA ( ▲ 0.04% ) ), (~1 million shares, ~$187 m) and Palantir Technologies ($PLTR ( ▲ 1.65% ) ) (~5 million shares, ~$912 m) for the quarter ending Sept 30.
• Nvidia’s stock has surged on the back of the AI craze this year. Palantir has jumped even more amid interest in AI and government contracts.
• Burry also posted on X: “Sometimes we see bubbles. Sometimes the only winning move is not to play.”
• For founders and business leaders this signals caution: when someone who bet big on a collapse makes a huge short in today’s top companies, it may reflect concern about frothy valuations.
• If you’re building a startup in an over-hyped sector, it might be time to ask: what happens when the market sentiment changes?
• This move isn’t a casual hedge; it shows conviction. It suggests risk in riding the wave without prepping for the reversal.
Takeaway:
When even the “Big Short” investor takes a step back, it’s time to seriously inspect the foundation beneath your growth.
Kimi K2 Thinking: The Open Model That’s Challenging GPT-5
Alibaba-backed Moonshot AI just made a bold move its new model, Kimi K2 Thinking, is taking on GPT-5, Claude Sonnet 4.5, and Grok 4 head-to-head.
Kimi scores high on agentic benchmarks
Highlights:
Kimi K2 Thinking posts competitive benchmark scores:
• 44.9% on Humanity’s Last Exam
• 60.2% on BrowseComp
• 81.1% on Real-World Tool Collection
• 61.1% on SWE-Multilingual
• 37.2% on SWE-Bench Verified
• 60.8% on LiveCodeBench v5
On several of these tests, it even surpasses GPT-5 and Claude Sonnet 4.5 (Thinking mode).
💡 The model uses a Reasoning Mixture-of-Experts (MoE) design with 1 trillion parameters and 32 billion active per token.
This design scales capacity without driving up compute costs; making K2 Thinking around 6× cheaper to run.
⚙️ It supports test-time scaling, spending more “thinking tokens” and tool calls on hard problems.
That means it plans, verifies, and revises its answers autonomously- much like a human expert checking their own work.
🧠 Interleaved thinking lets it pause between actions, reflect, and then continue with context intact.
It can read a document, run a search or code execution tool, then reason again with the results across hundreds of steps.
🗂️ A 256K context window allows handling long documents, multi-turn chats, or many tool outputs at once.
Focus shifts dynamically as the model refines its plan.
⚡ Serving is optimized with INT4 Quantization Aware Training (QAT), doubling generation speed while maintaining accuracy.
🪶 K2 Thinking is available on Hugging Face under a Modified MIT License that allows full commercial use and derivatives.
Only large-scale deployments (over 100 million users or $20 million monthly revenue) must credit “Kimi K2” in their interface.
So what?
Moonshot AI is rewriting the rules for open models. Kimi K2 Thinking proves that frontier-level reasoning and open access can coexist, and it might just be the moment when the AI giants get their first real open-source rival.
Reply