Good morning. It’s Wednesday, June 17th.

You read. We listen. Let us know what you think by replying to this email.

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

OpenAI readies bidirectional voice upgrade with new ‘GPT-Bidi’ architecture

OpenAI is prepping an upgrade for ChatGPT Voice with a bidirectional audio model called GPT-Bidi-1. Spotted across web and mobile frameworks, this architecture listens and speaks simultaneously, allowing the assistant to absorb user interruptions and adjust mid-sentence without the usual freezing. Designed to close the gap between OpenAI’s advanced text models and its older voice stack, it introduces High, Medium, and Instant intelligence tiers, letting users explicitly trade reasoning depth for lower latency.

This front-end push comes at a volatile moment. OpenAI burned $34 billion over the past year, funneling $19 billion into R&D and $6 billion into marketing, against a $13 billion revenue run rate. While paper losses hit $39 billion, $30 billion of that was a one-time accounting adjustment from corporate restructuring, leaving operational losses at $8 billion as the company preps for a $1 trillion IPO. Meanwhile, ChatGPT’s market share just slipped below the 50% mark for the first time, landing at 46.4%.

To protect its moat, OpenAI is building an OS and browser automation engine for Codex driven by three distinct interfaces. To bypass missing APIs, a Computer Use tool mimics human clicks and keystrokes across legacy desktop apps. For web workflows, a custom Chrome extension scrapes data from active, authenticated dashboards like Gmail, while an isolated, cookie-free in-app browser handles sandboxed web debugging. Codex glues this entire pipeline together using Appshots, a visual context stream that converts raw screen activity into automated workflows.

OpenAI is leaning heavily into OS and browser automation, detailing the three distinct interfaces driving the Codex ecosystem. The macro-level tool is Computer Use, which mimics human clicks and keystrokes to control legacy macOS and Windows apps that lack clean API integrations. For web tasks, a dedicated Chrome extension taps into active, authenticated browser sessions to scrape context from signed-in dashboards like Gmail or Salesforce.Then there is an isolated, cookie-free in-app browser designed specifically for sandbox web debugging and code modification. The glue holding this entire pipeline together is "Appshots," a visual context that captures that feed directly into Codex to turn raw on-screen activity into automated workflows.

On the backend, OpenAI introduced Deployment Simulation, running 1.3 million historical, de-identified user conversations through next-gen systems like the GPT-5-series. The testing pipeline successfully uncovered hidden vulnerabilities like "calculator hacking" while stripping out "evaluation awareness," the phenomenon where a model alters its behavior because it detects it is being benchmarked.

Anthropic delays Claude token price hike to survive OpenAI price war

Anthropic pulled back a billing overhaul for the Claude Agent SDK on the very day it was set to go live. The shelved update would have separated automated, programmatic tools like the Agent SDK and claude -p from standard flat-rate subscription pools, forcing heavy users onto usage-based API pricing once a small monthly credit ran out. Developers dodged a massive price hike for automation-heavy workflows, as Anthropic backed down to avoid a mass customer exodus.

The timing is critical: Anthropic is heading toward an IPO, and with OpenAI reportedly prepping steep API price cuts, a defensive developer-pricing standoff is taking shape. At the same time, Anthropic is locked in a high-stakes friction with the US Trump administration with an abrupt export ban on their flagship Claude Fable 5 and Mythos 5 systems. Even though Anthropic and independent security leaders argue the technical risks are overstated, the sudden blackout has triggered a major AI sovereignty crisis in Europe. European researchers are now deeply divided between heavily investing in local hardware or pursuing trade agreements.

This geopolitical chokehold is creating an immediate opening for open-source AI, specifically cheaper alternatives from Chinese firms like DeepSeek, Moonshot AI, and Z.ai. Displaced developers and governments looking to escape U.S. regulatory whiplash are shifting to these locally deployable models, which are already dominating top usage rankings on platforms like OpenRouter.

Despite the regulatory chaos, Anthropic's underlying tech remains elite, with Claude Fable 5 hitting a top score of 161 on the Epoch Capabilities Index and outpacing GPT-5.5 Pro on complex engineering benchmarks.

Frontier models and product moves

Agents and the agentic stack

Business, labor, and institutions

Security

Hardware and infrastructure

Research and benchmarks

Robotics

Media and creative AI

Watch

NoteRich offers a private and powerful note-taking hub with local storage and secure AI features.

Innflow connects your favorite tools to deploy AI agents that handle repetitive tasks automatically.

MakersClaw deploys autonomous, specialized AI employees directly into Slack, Teams, and Telegram to handle business tasks.

Invoko is an AI desktop assistant that executes tasks and drafts content through voice-activated context awareness.

Novu Connect integrates Claude agents into communication channels like Slack, WhatsApp, and email for automated workflows.

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!

Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.