Welcome, humans.

Remember when Fable 5 was such a big deal the U.S. government briefly hit pause on it? Yeah, that Fable 5. Anthropic's smartest model ever, unveiled to genuine hype back in June.

A month later, businesses have made their verdict: meh. In Ramp’s August 2026 AI Index, Fable 5 makes up only 6% of the tokens businesses buy from Anthropic, and just 11.4% of their spend, despite costing roughly double GPT-5.6 Sol per token. OpenAI's "boring" cheaper model is still generating more total business spend than Fable 5, the actual smartest model on the planet.

Fable 5 (the dark purple bars at the very bottom) barely registers next to everything else businesses are actually paying for.

Meanwhile, Sam Altman said AI probably won’t give us the shorter workweek because humans apparently enjoy staying busy. Then Google Search briefly declared him dead. Google really took “clear his calendar” literally. 😹

Here’s what happened in AI today:

😸 Google, OpenAI, and DeepSeek all dropped new AI models within 24 hours.
📰 Anthropic found its AI agents sabotaged each other with self-replicating malware.
📰 OpenAI lost its second senior executive in a week.
📰 Microsoft quietly closed 15+ offices in China over five years.
🎓 Turn an AI-generated Marble world into a walkable Unreal Engine scene.

😺 Google, OpenAI, and DeepSeek All Dropped New Models on the Same Day

If you stepped away from your inbox yesterday, you missed a pileup. Three of the biggest names in AI (Google, OpenAI, and DeepSeek) each shipped a model update within about 24 hours, and each picked a different way to compete: price, speed, or flexibility.

Here's what happened:

Google launched Gemini 3.7 Flash, its go-to "workhorse" model for coding and AI agents (software that can complete multi-step tasks on its own), at half the price of its predecessor.
OpenAI began previewing Ultrafast, a new setting that runs its GPT-5.6 Sol model up to 14 times faster, generating up to 750 tokens (small chunks of text) per second.
DeepSeek shipped V4-Pro, adding adjustable "thinking" levels so developers can dial reasoning up or down depending on the task, plus cheaper off-peak pricing.

The gains are real, not just marketing. Gemini 3.7 Flash jumped from writing correct, production-ready code 34% of the time to 44% on one coding benchmark (a standardized test for comparing AI models), and doubled its score, from 22% to 34%, on a test for reading dense documents like financial reports.

OpenAI's Ultrafast mode runs on chips from Cerebras and is already being tested by firms like Jane Street and Podium for jobs like fraud detection and live customer support, where delay costs real money. DeepSeek's new pricing kicks in Sunday, with off-peak rates running 50% cheaper than peak hours.

Why this matters: We're watching a shift from smartest to cheapest. The smartest model in the room used to win, full stop. Fast forward to 2026, with more people and companies running AI daily, and the smartest pick isn't always the smartest, resource-wise, since smarter models burn more tokens (how usage gets billed). Cheaper models mean more compute to go around, while the smartest stay reserved for tasks that need them. AI now drives so much automation that cheap just became the priority.

FROM OUR PARTNERS

When WHOOP launched their most ambitious product yet, they didn't just want faster ticket resolution, they wanted to catch issues before they became tickets at all.

With Unwrap’s AI customer intelligence, that's exactly what happened:

50% faster root cause investigation on support tickets
4X drop in launch-day ticket spikes vs. their previous launch
24+ hours average saved getting ticket volumes back to baseline

If your team still sorts through customer feedback manually (or with a patchy mix of manual work and AI), there's a better way. Unwrap brings all your feedback across surveys, reviews, tickets, social comments etc. into one view, then uses AI and NLP to surface what actually needs your attention.

As a subscriber of The Neuron you can try Unwrap for free and join teams like WHOOP, Perplexity, Stripe, and DoorDash using it at scale.

Try Unwrap for free

🎓 AI Skill of the Day: Turn a Marble World Into an Unreal Scene

World Labs’ 18-minute tutorial shows how to turn a Marble-generated world into a walkable Unreal Engine scene using the VIVE Mars Nova 3DGS plugin.

Download the Marble environment as an SPZ using OpenGL coordinates, plus its GLB collider and 360 panorama. Convert the panorama to HDR, install the matching VIVE plugin in your Unreal project, then import all three. Attach the collider to the splat blueprint, set collision to “Use Complex Collision as Simple,” hide the collider, and use the HDR as your skylight.

Watch the full tutorial for the plugin setup, lighting workflow, and clipping fix.

Have a specific skill you want to learn? Request it here.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

*Adobe Podcast cleans noisy recordings, removes echo, and lets you edit speech from a transcript so rough audio sounds studio-ready; free plan available.
Lettertrace tracks how ChatGPT, Claude, and Gemini describe your brand so you can monitor visibility, sentiment, and competitors over time.
Dograh gives you a self-hostable voice-agent stack with telephony, human handoff, model choice, and MCP support.
mcptoon keeps MCP schemas on disk and serves agents compact tool formats to cut discovery and result-token overhead.
Nuphos gives your DevOps team AI agents that can investigate incidents, hunt down idle cloud spend, and plan migrations, with permissions and audit trails built in —free to try, agents from $29/mo.
Ito builds and runs your actual app on every pull request, catching bugs a code diff alone can't see and dropping video proof right in the PR —free for your first 100 reviews, then $40/mo.
mcp-memory gives AI coding tools like Claude, Cursor, and Codex a memory that survives between sessions, so your agent picks up right where it left off —free to try.

📰 Around the Horn

Anthropic just gave every word Claude writes a permanent paper trail. Copy it, paste it, edit it all you want: the receipts don't wash out.

Anthropic found that when it set several Claude agents loose on the same project without telling them about each other, they sabotaged each other with self-replicating malware before eventually negotiating a truce.
OpenAI lost its second senior executive in days, as chief revenue officer Denise Dresser departs less than a year into the job ahead of the company's expected IPO.
Microsoft has quietly shut over 15 offices and joint ventures in China over the past five years, even as it keeps a foothold there for its AI and cloud business.
OpenAI launched Computer History, an opt-in feature that lets ChatGPT recall what you did on your Mac across apps and sites without taking constant screenshots.
GenBio AI unveiled a "virtual cell," a digital simulator that lets researchers test drug and gene-editing ideas on a computer before running the real experiments.
In a fun tangent, a new preprint proved the 60-year-old Crouzeix conjecture in matrix math, nailing down the optimal constant mathematicians have chased for decades.

FROM OUR PARTNERS

Secondary ad here

See Why HubSpot Chose Mintlify for Docs

HubSpot switched to Mintlify and saw 3x faster builds with 50% fewer eng resources. Beautiful, AI-native documentation that scales with your product — no custom infrastructure required.

Simplify Your Docs Today

📖 Intelligent Insights

Every frontier model here basically flunks a coin flip until you write the prompt like an expert then it jumps 30 points and still can't crack 80%.

[Rotating section image goes here]

Anthropic reviewed 56 randomized U.S. retraining studies and found typical programs improved employment only 2-3 points, raising questions about readiness for AI displacement.
Thinking Machines Lab found an expert-fine-tuned Qwen3-235B model beat every frontier model tested on financial filtering while costing 13.8x less to run.
OpenAI says frontier firms now generate 8.3x more output tokens per active user than typical enterprises, suggesting the AI gap is depth of use, not access.
AT&T says open-weight models already power about 25% of its AI usage, helping control token costs and keep proprietary data in-house.
Halluminate's Westworld benchmark put Opus 5 first at just 0.51 average score across an end-to-end acquisition workflow, showing how hard long-horizon finance remains.