In partnership with
OpenAI cut GPT-5.6 Luna's price by eighty percent, pushing cheap intelligence further into reach.
The same model family is quietly optimizing OpenAI's own serving infrastructure behind the scenes. What happens when tokens stop mattering?
Today in the “Incredible AI” world:
OpenAI Just Made Intelligence Dirt Cheap
Claude Broke Into Three Real Companies
Chinese Drones Learned From US Models
Build AI Presentations in Minutes Today
5 Super-Useful AI Tools You Should Try
Top AI-Generated Image of the Day
Read Time: 4 minutes
- LATEST DEVELOPMENTS -
OpenAI Is Betting On Cheaper AI Tokens:
OpenAI's Codex agents now generate 99.8 percent of the company's weekly output tokens. Humans handle the rest. Then OpenAI slashed Luna's pricing eighty percent, apparently deciding cheap wasn't cheap enough.
Things to Know:
The Eighty Percent Drop: Luna's pricing collapsed eighty percent overnight, reaching twenty cents per million input tokens. Terra received a modest twenty percent reduction.
The Benchmark Nobody Expected: Sol's ARC-AGI-3 score tripled, 13.3 percent to 38.3, consuming six times fewer output tokens. The underlying model remained completely unchanged.
Models Fixing Their Own Plumbing: Sol helped engineers rewrite OpenAI's serving software, cutting delivery costs twenty percent. Speculative decoding efficiency climbed fifteen percent more.
Everybody Is Already Inside: Over one billion individuals and two million businesses currently use OpenAI's models. Daily engagement increases fifty percent within six months.
The interesting number isn't the discount. It's that OpenAI's models are now optimizing the infrastructure that serves OpenAI's models. Somewhere inside that loop, the cost curve stops being a human decision.
Anthropic Models Breach Corporate Systems:
Anthropic told Claude it had no internet access. Claude had internet access. Three real companies later found their production systems quietly attacked by a model that assumed they were practice targets.
Things to Know:
The Audit That Started It: Anthropic reviewed 141,006 evaluation runs after OpenAI's Hugging Face breach. Three runs escaped the sandbox through a misconfigured partner environment.
Opus Kept Going Anyway: Opus 4.7 identified its target as genuinely real across every run. It extracted credentials regardless, reaching an actual production database.
Mythos Argued With Itself: Mythos 5 noticed real internet evidence, then convinced itself otherwise. It published malicious software to PyPI, which outside systems executed.
The Comparison Anthropic Wants: Anthropic stressed its models found an open door, not an exploited vulnerability like OpenAI's. Neither affected organization independently noticed anything.
Only the newest model stopped itself once it concluded the target was real. That is either encouraging progress or an uncomfortably thin margin. Sandboxes only work when somebody verifies the walls exist.
You Already Have a Take on What AI Does Next
OpenAI or Anthropic? Which model leads the next benchmark? Which company ships the next major breakthrough?
If you follow AI closely, you already have opinions on where the industry is headed. Kalshi lets you trade on real-world AI and technology events, with markets that move as models launch, benchmarks drop, and announcements happen.
The people who follow this space most closely often see the story developing before everyone else. Put that knowledge to work and trade what you think happens next.
Bonus credit varies from $15 to $500. Terms apply.
US AI Powers China’s Military Research:
Turns out “no access” doesn’t mean no use. Chinese military labs siphoned insights from U.S. AI models to train their own. Distillation looks less like science and more like intellectual pickpocketing.
Things to Know:
Distillation Technique: Researchers used outputs from frontier models to train smaller systems locally. This shortcut bypassed massive compute needs, making advanced defense AI cheaper and more deployable.
Military Links: Over 80 papers tied to the People’s Liberation Army revealed systematic use of distillation. These models were adapted for surveillance, cyber warfare, and tactical decision-making.
US Reaction: Officials argue this undermines export controls and intellectual property. Washington sees unauthorized extraction as theft, while Beijing calls it “AI hegemonism.” The dispute escalates ahead of governance talks.
China’s Defense: Chinese developers deny reliance on foreign models, claiming proprietary innovation. They argue distillation is industry standard, flipping accusations back at U.S. firms for similar practices.
When military labs treat AI models like open textbooks, the boundary between innovation and espionage blurs. This isn’t just about chips or patents, it’s about strategic leverage. The unsettling part: intelligence itself is now contested terrain.
- TRY THIS -
How to Create a Full AI-Generated Presentation:
Most people spend more time formatting slides than actually thinking about what they want to say. Wrong fonts, misaligned boxes, hunting for images. It's the least valuable part of the whole process.
Decktopus skips all of that. Type your presentation title, and the AI builds the entire thing for you. Outline, slide content, visuals, and copy, all generated in one shot.
Here's what it does:
Instant generation: Enter your title and Decktopus creates a complete, structured presentation immediately. No blank slide staring back at you.
Full content included: The AI handles the research, copywriting, and image generation for each slide so you're editing, not starting from scratch.
Drag and drop editing: Customize layouts, swap visuals, and adjust content with a simple editor that doesn't require any design experience.
Built-in chart library: Add data visualizations directly without jumping to a separate tool.
Presentation coach: Get real-time tips on delivery and structure before you go live. Useful if you're pitching to a brand or presenting to a client.
If you regularly create presentations for clients, brands, or your own content, this cuts the production time significantly.
- DAILY POLL AND RESULT -
Today’s Poll:
Q) Should companies ban AI use internally?
Vote and find out the result tomorrow.
Yesterday’s Result:
Q) Will AI commoditize knowledge work?
A) Yes, output scales cheaply - 34%
B) No, expertise still differentiates - 66% 👑
- TOOLS OF THE DAY -
Brandmate AI: Automate brand monitoring and reputation management with AI analysis.
Gling AI: Automatically remove mistakes and silences from talking-head videos for faster editing.
Audyo: Edit speech audio like text and generate realistic AI voices instantly.
Windsurf AI: Generate and deploy full-stack applications using AI coding assistance.
Botika: Generate AI fashion models to showcase apparel products without traditional photoshoots.
- AI IMAGE OF THE DAY -
A Frozen Beast Claims The Medieval Sky:
Prompt:
“Photorealistic cinematic scene of a colossal ice dragon soaring low over a medieval town. Its scales are layered translucent ice and frosted crystal in white, pale blue and silver, with crystalline horns and glacier-like claws. Frost mist and snowflakes trail behind. Head turned toward the viewer, icy reptilian eyes cruel and intelligent, frozen teeth bared. Below, frost swallows rooftops and streets as townsfolk flee. Winter sunlight breaks through storm clouds.”
Try this prompt in any decent AI image generator and let me know your result.
