• AI PlanetX
  • Posts
  • Reflection AI Enters Open Model Race

Reflection AI Enters Open Model Race

OpenAI Adds Invisible Watermark

In partnership with

AI PlanetX

Welcome to another edition of AI PlanetX.

Reflection AI enters open model race; ChatGPT adds invisible watermarks as Meta and Microsoft pull back from Claude.

Inside This Edition: đź’Ž

  • Hottest AI News

  • Top AI & SaaS Tools

  • AI Tutorial: Connect ChatGPT to Apple Messages

  • Top AI & Tech News

  • AI Art Spotlight

  • Prompt of the Day: Prompt That Makes Hermes Audit Itself

  • Featured AI Video

Hottest AI News

Reflection

Reflection AI Enters Open Model Race With Beam Built for Coding Agents

Reflection AI just entered the open-model race with Beam, an open-weight system for coding, reasoning and agents. The $25B U.S. startup founded by ex-DeepMind researchers is betting on efficiency as its edge.

Details:

  • Beam has 501B parameters, with 23B active at once. Reflection says it matches GLM-5.2 on several tests while using about 3–4x less inference compute. The benchmark remain self-reported

  • Beam still trails Kimi K3 and newer leaders like GLM-5.3 on many tests, but Reflection says it beats U.S. open models such as Inkling and approaches Qwen 3.8-Max on coding and agent tasks

  • Weights are due this month under Apache 2.0, so companies can customize and self-host Beam. Reflection is also building private AI infrastructure, including a 250MW sovereign AI cloud in Korea

Beam is not leading the open-model race yet, but it gives the U.S. a real contender. The bigger question is whether Reflection can narrow China’s lead while keeping Beam efficient.

Your SOC 2, handled end to end with AI agents.

Enterprise buyers won’t put your product near their customer data without a SOC 2 report. Twenty-one state privacy laws now hold them accountable for the vendors they share data with, so their diligence lands on you.

Sprinto gets you audit ready in 14 days, across three working sessions. AI agents connect your stack, collect the evidence auditors ask for, and close the gaps as they appear. You approve, they execute.

Sprinto also answers the security questionnaires your buyers send, using the same evidence base, so security reviews stop holding up deals.

No compliance hire. No consultant. Your auditor signs off.

OpenAI

ChatGPT Is Adding Invisible Watermarks to AI Generated Text in Europe

OpenAI is adding an invisible watermark to ChatGPT and Codex text in the EU. Called textGrain, it changes statistical word choices so software can detect AI output, helping OpenAI meet the EU AI Act’s machine-readable rules.

Details:

  • Over the coming weeks, ChatGPT and Codex users on every EU plan will get it. API customers worldwide can opt in for select models; it stays off by default, with no global rollout at launch

  • OpenAI says textGrain matched or beat SynthID for text with no benchmark drop. It plans to open-source the method, but the detector is not public: approved researchers and expert groups can apply

  • Detection is fragile: about 80% at 200 tokens vs 95% at 400, at 1% false positives, and worse for math. Replace 10% with synonyms: 92% falls to 66%. Replace 25%: just 17%. A miss proves nothing

This is useful for EU compliance, but not foolproof proof of authorship. A watermark cannot identify the user, measure human edits, or prove accuracy.

Top AI & SaaS Tools

  • Snappyit (Life-time Deal): AI-powered product photography platform that turns one garment photo into ecommerce images, color variants, and short videos

  • Clara AI: AI-powered SDR that converts website visitors into qualified sales opportunities through real-time, automated conversations [F-R-E-E to Try]

  • Similarweb AI Ads: Lets marketers track competitor ads, creatives, user intent, and share of voice across ChatGPT, Google AI Mode, and AI Overviews to plan and optimize AI advertising campaigns

  • OpenClaw 2.0: Simplifies setup, adds a first-class browser app and collaborative sessions, and expands automation, messaging, memory, models, plugins, and security

  • Kilo Code for JetBrains: Open-source, native coding agent for JetBrains IDEs, supporting local and remote development, parallel worktrees, GitHub integration, and 500+ AI models

Stop typing AI prompts. Start talking.

You think 4x faster than you type. So why are you typing prompts? Wispr Flow turns your voice into ready-to-paste text inside any AI tool. Speak naturally, tangents and all, and Flow cleans it up. Available on Mac, Windows, iPhone, and Android.

AI Tutorial

How to Connect ChatGPT to Apple Messages on Mac

Want ChatGPT to work with your iMessages? A built-in Messages integration lets ChatGPT interact with Apple Messages on your Mac, making it easier to catch up on conversations, find important texts, and draft responses.

  1. Open ChatGPT on Mac

    Launch the ChatGPT desktop app and sign in. The Messages integration requires a Mac with Apple silicon and works through ChatGPT Work or Codex.

  2. Find Messages Plugin

    Open Plugins, search for Apple Messages, and install it. Depending on your account, the interface may show plugin options differently.

  3. Grant Permissions

    Follow the setup prompts and allow the required Mac permissions:

    • Full Disk Permission to search messages

    • Contacts to identify people

    • Automation to work with Messages

  4. Start Work Chat

    Switch to Work and start a fresh chat. For example:

    “Check my recent unread messages and summarize anything I should respond to. Then draft a text to myself with the recap.”

    ChatGPT Work is available directly from the desktop app.

  5. Approve Message Permission

    When ChatGPT asks to see conversations, approve the messages you want it to use.

  6. Let ChatGPT Handle Task

    ChatGPT can search conversations, summarize messages, suggest follow-ups, and prepare replies. It supports iMessage, SMS, and RCS.

  7. Review and Send

    Check the generated message and recipient, then approve the send. Message sending requires user approval by default.

Note: Review generated replies before sending them, especially for important or sensitive conversations.

Top AI & Tech News

  • OpenAI is seeking $30 billion in latest funding, with UAE investment funds considering a syndicate of up to $10 billion to anchor the round

  • Meta and Microsoft are reducing their internal use of Anthropic’s Claude as they prioritize cheaper in-house and competing AI tools

  • Anthropic expanded its Claude for Startups program, offering eligible startups a free year of Claude Team, $1,000 in API credits, and additional support for building AI-powered products

  • LibreOffice will not include built-in AI features for the foreseeable future, prioritizing privacy, offline functionality, and user control over data

AI Art Spotlight

Create Your Own Photo Like This Using the Prompt Below:

Premium commercial food advertising photograph of [Main Dish] resting naturally on a rustic wooden table, hero dish fully visible and sharply focused, [Floating Ingredients] suspended above and around the dish, individual pieces clearly separated, a controlled arc of [Sauce/Liquid] splashing through the air behind the food, droplets frozen in motion, realistic appetizing textures, directional studio lighting, shallow depth of field, uncluttered composition, generous empty space for advertising copy, high-speed food photography, photorealistic

Model: ChatGPT Images 2.5 / Nano Banana 2

Prompt of the Day

The Prompt That Makes Hermes Audit Itself

This prompt makes Hermes check its own setup and show you what could be improved. It reviews Hermes Doctor and Hermes Dump, finds the three most important issues based on how you use it, explains each one in simple words, and gives you the exact command to fix it.

Featured AI Video

Write VIRAL AI Video Prompts with NotebookLM (F-R-E-E & Unlimited)