Good morning, AI enthusiasts. OpenAI wants to be the AI backbone of the U.S. government, and it’s going all in to undercut rivals to get there.

With a new initiative offering ChatGPT Enterprise to federal agencies for just $1 per agency, the AI giant is making a bold push for long-term government adoption, but it’s also one that could change expectations around enterprise pricing.

Reminder: Our next live workshop is tomorrow at 4 PM EST. Join and learn how to run OpenAI’s new open weight models locally on your computer. RSVP here.

In today’s AI rundown:

• OpenAI offers ChatGPT at $1 for U.S. agencies

• Google launches AI tutoring mode for students

• Create Claude Code subagents for your projects

• Microsoft’s self-adapting AI tackles scientific problems

• 4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

OPENAI

🇺🇸 OpenAI offers ChatGPT at $1 for U.S. agencies

Image source: The Rundown

The Rundown: OpenAI just announced that the enterprise version of ChatGPT, including access to advanced models with additional security features, will be available to all federal agencies for just $1 per agency for the next year.

The details:

• The discounted pricing stems from OpenAI's partnership with the U.S. General Services Administration, the government's central purchasing authority.

• It will also be accompanied by unlimited use of OpenAI’s advanced models and features, like Deep Research, for an additional 60-day period.

• The company hopes these efforts will help government officials cut down red tape and paperwork, making services “faster, easier, and more reliable.”

• To help federal employees get started, OpenAI is also setting up a dedicated government user community with tailored training resources.

Why it matters: While offering ChatGPT Enterprise for virtually free shows OpenAI’s willingness to embed itself deep into government workflows, it could also spark a high-stakes race, with other GSA-approved rivals like Anthropic and Google likely following suit and unleashing a wave of aggressive AI offers aimed at federal adoption.

TOGETHER WITH POSTMAN

🏆️ Build something incredible to win

The Rundown: Turn your API expertise into $1,000 and prizes in Postman's newest challenge. Create interactive Notebooks to showcase AI-powered solutions. After all, AI + APIs is a match made in heaven.

The prizes include:

• $1,000 for multiple categories

• $2,500 + for Creator Fellowship Grant

• $1,000 for Community Favorite and more

Join the challenge and submit your work by August 24 to be considered.

GOOGLE

🧑‍🎓 Google launches AI tutoring mode for students

Image source: OpenAI

The Rundown: As OpenAI courts federal agencies with $1 licenses, Google is pushing into the education sector with a new Guided Learning mode for Gemini and free access to its $250/month AI Pro Plan for college students.

The details:

• Following ChatGPT’s Study Mode, Gemini’s Guided Learning acts as a learning partner, offering step-by-step guidance instead of giving direct answers.

• Google has worked with educators and learning experts to ensure the AI helps students work through problems, building their critical thinking skills.

• The mode also includes tools that use images, videos, and interactive quizzes to help students test their knowledge as they learn new concepts.

• Google is making its AI Pro Plan free for students in select countries, including the U.S., and investing $1B over 3 years for AI training at U.S. colleges.

Why it matters: Amid concerns that AI may hurt the learning process with direct answers—including an MIT study highlighting its impact on students’ cognition—both Google and OpenAI are now repositioning their AI tools with study tutor features designed to build, not bypass, critical thinking.

AI TRAINING

🤖 Create Claude Code subagents for your projects

The Rundown: In this tutorial, you’ll learn how to use Claude Code's sub-agents to create custom AI assistants with specific roles and instructions, each with their own context window and expertise.

Step-by-step:

1. Install Claude Code with npm install -g @anthropic-ai/claude-code and run claude in your project directory

2. Type /agents, select “Create a new agent” and choose “Generate with Claude”

3. Describe your agent's purpose and configure tools, model (Opus/Sonnet/Haiku), and color theme

4. Save your agent, and it will automatically activate for relevant tasks.

Pro tip: Create multiple specialized sub-agents (testing, reviewing, bug-fixing) to maintain focused expertise and separate context windows for each task.

PRESENTED BY TOGETHER AI

⚙️ Automate, debug, and solve with SOTA open source models

The Rundown: Together AI's Inference & Fine-Tuning platform lets developers deploy open source models at scale, offering both serverless and dedicated endpoints for models like Kimi K2 — a new SOTA agentic intelligence model with 1T parameters, 128K context, and native tool use.

With Together AI, you’ll experience:

• High-performance infrastructure designed for the lowest possible latency

• Developer-friendly APIs that simplify integration

• An extensive model library featuring the top open-source AI models like Kimi K2

Try Kimi K2 now and get all the benefits of open-source AI without the complexity.

AI RESEARCH

🧠 Microsoft’s self-adapting AI tackles scientific problems

Image source: Microsoft

The Rundown: Microsoft just announced CLIO (Cognitive Loop via In-situ Optimization), a breakthrough framework that enables non-reasoning LLMs to develop their own thought patterns and adapt their reasoning in real-time.

The details:

• While most reasoning models use strategies and actions that are pre-baked during post-training (before deployment), CLIO creates a “steerable” AI system.

• It builds and refines its reasoning through self-reflection at runtime, creating its own feedback loops to explore ideas, manage memory, and flag uncertainties.

• With the self-adapting behavior, users get full control to set thresholds for uncertainty, change reasoning paths, or re-execute them entirely.

• On Humanity’s Last Exam, CLIO boosted GPT-4.1’s accuracy on text-only biomedical questions from 8.55% to 22.37%, surpassing o3 (high).

Why it matters: With CLIO’s performance leap—plus built-in explainability, memory control, and tunable reasoning—Microsoft has shown that LLMs don’t need to be “finished” in training. In high-stakes scientific domains where trust and rigor matter, a continually steerable AI might be just what teams need to accelerate discovery.

QUICK HITS

🛠️ Trending AI Tools

• ✨ Anything - Launch beautifully designed mobile apps and web apps from just a prompt, no coding or extra tools required*

• ⚙️ Bing Image Creator - Now with OpenAI’s latest GPT-4o image model

• 📽️ Endex - An AI agent that lives inside Excel

• 📝 Kitten TTS - A 15M parameter text-to-speech model that runs without GPU

^{*Sponsored listing}

💼 AI Job Opportunities

• 🧠 Meta - AI Research Scientist, Computer Vision

• 🎨 Descript - Product Designer

• 🤝 Deepmind - People Partner, GenAI

• 📈 UiPath - Territory Sales Manager

📰 Everything else in AI today

Midjourney added a new HD video mode for its Pro and Mega subscribers, saying it delivers 4x more pixels in outputs than the default SD mode and is 3.2x more expensive.

Google’s asynchronous, agent-based coding tool, Jules, is finally out of beta, but the free plan usage limits have been reduced to 15 individual daily tasks from 60.

Anthropic added automated security reviews in Claude Code with a new /security-review command that checks for vulnerabilities directly from the terminal.

xAI’s Grok Imagine, a video and image generation tool, is now rolling out in early access on Android devices, via the official Grok app.

OpenAI announced a “LIVE5TREAM” for Thursday, 10 am PT (today), teasing the launch of GPT-5, its long-awaited flagship model, with possibly a Nano and Mini version.

OpenAI is also reportedly holding early talks about a secondary share sale for current and former employees at a valuation of $500B.

Chai Discovery, the AI lab building frontier models for molecular design, raised $70M series A from Menlo Ventures and the Anthology Fund.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop this Friday, August 8th, at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll know how to run your own fully private, open-weight model locally on your computer.

RSVP here. Not a member? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer—the humans behind The Rundown