In partnership with
🧠 THAT ONE AI - This Week’s Signal
Here’s what’s shaping AI right now - without the noise:
🎨 🫠 GPT-5.6 Sol is here but almost nobody can use it
🧠 One AI tip worth applying today
🇨🇳 Chinese models are closing the gap faster than expected
🧰 Tools worth testing
🫠 GPT-5.6 Sol Is OpenAI's Best Model and You Probably Can't Access It
OpenAI just previewed GPT-5.6 Sol, its most capable model to date. The catch: at the U.S. government's request, access is currently limited to around 20 vetted partners.
The 5.6 family comes in three tiers:
Sol > the flagship, with max reasoning effort and an ultra mode that spawns parallel subagents for complex tasks
Terra > the balanced option, matching GPT-5.5 performance at roughly half the cost
Luna > the fastest and cheapest of the three
On benchmarks, Sol beats Mythos 5 on Terminal-Bench 2.1 and matches it on ExploitBench while using about a third of the output tokens. That efficiency gap is worth paying attention to.
One honest note: METR's evaluation found Sol cheating on its own evals at a higher rate than any previously tested model. OpenAI says protections were trained in. The results say those protections are not complete.
OpenAI has acknowledged that government-gating should not be the long-term default. But if this becomes the pattern for frontier releases, most builders will consistently find themselves waiting while a small group of vetted partners gets first access.
The bigger signal: The frontier is bifurcating. Government-controlled access at the top, open competition below it. How that gap evolves will shape what most developers can actually build with.
Wake Up Smarter About AI.
Most AI news is a waste of time. The Future Today is a daily 5-minute read focused on what actually matters.
You’ll get exclusive interviews with the CEOs, researchers, and builders shaping AI. Plus top stories broken down simply and practical advice you can immediately apply.
Read the newsletter trusted by teams at NVIDIA, Google, Anthropic, Meta, Dell, and Salesforce.
🇨🇳 Chinese Models Are Closing the Gap On Price and Performance
Two developments this week make it harder to dismiss Chinese AI progress as a benchmark story.
360 Security Technology launched Tulongfeng, a model its founder described as a Chinese equivalent of Mythos for identifying cybersecurity vulnerabilities. Separately, Z.ai released GLM-5.2, which rivals top US model performance at a significantly lower price point.
The cost angle is what makes GLM-5.2 particularly worth tracking. Frontier-level performance at a fraction of the price is not just a win for Chinese labs it creates pricing pressure on every US lab competing for the same enterprise customers. Read more from WSJ.
Neither model has been independently verified at the levels claimed. But the pattern is consistent enough now that treating Chinese AI as a lagging competitor is no longer a safe assumption.
The bigger signal: The performance gap between US and Chinese frontier models is narrowing. The price gap may already be gone.
🧠 That 1 AI Tip: How Anyone Can Spot AI-Written Content
Most people assume you need a detection tool to catch AI-generated writing. You do not. Once you know what to look for, the patterns are obvious.
A few tells that show up constantly:
Bullet points that should be prose. AI defaults to lists even when an argument would read better as flowing text. If every idea arrives as a tidy three-point chunk, that is a mechanical habit, not a structural choice.
The causal pivot, overused. Phrases like "this is not just a product launch it is a strategic inflection point" are designed to add drama. AI reaches for this construction constantly. One causal pivot can land well. Five in the same piece signals automation.
Triplets everywhere. AI loves the rule of three. Beginning, middle, end. Speed, scale, security. Problem, solution, outcome. When every paragraph contains exactly three examples, the rhythm becomes predictable in a way human writing rarely is.
Sterile precision. AI writing is often technically clean and completely soulless. No stray metaphors, no rhythm changes, no moments where the writer seems to have an actual opinion. If a piece has no friction at all, it was probably not written by a person.
Why this matters for your own writing: If you use AI to draft content, these are the patterns editors and readers will notice first. The fix is not to avoid AI. It is to treat the output as a first draft, inject a real perspective, break the rhythm occasionally, and remove the rhetorical inflation before it goes out.
AI is the intern. You are the editor. The content should reflect that.
Pro Tip: consider using Humanizer
That One AI 🧰 TOOLBOX
A few tools quietly worth exploring:
📚 Google Study Notebooks → Personalized learning and study tools now built directly into the Gemini app
💻 Codex → OpenAI's coding tool is now generally available inside ChatGPT
📽️ ByThen → Turn any idea into a ready-to-publish faceless video with AI
🌐 Durable → Build a professional website in 30 seconds with AI
🔚 Before you go > next week's AI tip is your call.
What do you want us to cover for That 1 AI Tip?
See you next issue.