I'm unable to make up my mind on what to write today.

Anthropic shipped Opus 5.5 on Tuesday and OpenAI replied the same day with two cheaper GPT-6 models.

A day later, the heads of both labs were at the UN asking for guardrails. Meanwhile Claude found something weird in viral DNA, and Meta put its AI agent on a keychain.

Pick one? I couldn't.

So I went back to July. When Opus 5 launched, reviewers called it neurotic, so I tested it myself. Now Anthropic says 5.5 explains things more clearly, puts the important stuff first and cuts the jargon.

Big claim. So I put both in the ring. But first, the catch-up.

GPT-6 got cheaper. Sol and Luna cost half of what they replace. Catch: for now they only live in Codex and ChatGPT Work, so you'll have to go looking.

ChatGPT Voice can use your apps. It now pulls from your email, calendar and Slack, so "what does tomorrow look like?" finally gets an answer.

The AI bosses went to the UN. Altman and Amodei briefed the Security Council on Wednesday. Altman's headline: "We could lose control of the future to AI." This was a day after both made their AI cheaper for everyone. Great timing. 😅

Meta went all in on Muse. It can now use apps on your Mac and shop for you, and it's getting its own email address. Can't wait? Instinct already does this over iMessage and WhatsApp (invite-only though).

Claude found an enzyme system nobody had seen before. Claude agents were digging through viral DNA and spotted CRISPR-like repeats in the raw data. Nobody knows what it does yet, and 10 reruns of the same search missed it completely. (hold onto that one)

Describe a voice in plain words and it builds it. 2,000+ voices, 100+ languages, and it can clone yours from a 30-second clip. Already in Gemini Notebook.

A Muse-style agent you host yourself. It gets a browser, your files, Gmail and Calendar, and since it's open source, your inbox never goes through Meta.

Point it at a page or topic and it only pings you when the thing you care about changes. The free launch window just ended, so it's pay per check now.

Two fresh chats, memory off, default effort, same prompts.

Prompt: I have 20 minutes before a meeting. Explain quantum computing to me so I can explain it to a 15-year-old. I don't want a textbook explanation. Give me one analogy I can actually remember, then tell me exactly where the analogy breaks. All within 350 words max.

Both picked the same analogy: noise-cancelling headphones. Wrong answers cancel out like noise, and the right one is what's left.

Then they split.

Opus 5 talked like a friend. A normal computer walks a maze one path at a time, and a quantum computer "sends a wave into the maze." 

It also closed with the best line in either answer: if the kid asks how the cancelling works, just say "I don't know, that's the hard part."

5.5 handed me a textbook chapter, with a big title, headings, and qubit, phase and entanglement in the first few lines. 

Literally what I asked it not to do. 🙃 It was more complete though, and stuck closer to my word limit (Opus 5 went about 30 over).

Prompt: That analogy makes quantum computing sound like trying lots of possibilities at once. Isn't that misleading? … Explain what's actually different without using another analogy. All within 200 words max.

Bit unfair of me, since both had already said quantum computers don't try every answer at once.

Opus 5 took the blame anyway ("the fault was mine"), then lost me with "one reversible linear operation to the entire amplitude vector." Try saying that to a 15-year-old.

5.5 agreed with the fair part, reminded me it had already covered this, and kept it plain: a normal computer's odds can only add up, while a quantum computer's numbers can go negative and cancel out.

The bill: Opus 5 cost $0.072 and 5.5 cost $0.055, about a quarter cheaper. Token counts were nearly identical, so that's mostly the price cut.

I went in expecting 5.5 to sweep but it didn't.

Opus 5, the model everyone called annoying, still writes the warmer first draft and the line I'd use. But push it once and it apologises and starts lecturing, while 5.5 gets calmer and clearer.

So maybe the upgrade is a better second answer.

In July I said the fight had moved to price, speed and tone. 5.5 wins on price, but on tone, I’m split. And one run each proves very little.

Remember the DNA find? Spotted once, missed ten times in a row.

Your homework: ask both the same question, push back once, and see which one you'd trust with your second message. Then reply and tell me which explanation you'd have used.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Teams like Parallel, Turbopuffer, and Wordsmith are already setting the pace on Attio. Get an always-on revenue engine, with agents and workflows that build pipeline, chase every buying signal, and move deals forward with your team. Whether you're working in your browser, inbox, or favorite agent, connect to your customer data in real-time through Attio's web app, MCP, API, and SDK.

Try Attio now

#AD 2

Engagement is easy to mistake for pipeline. On Sep 30, watch how a founder turns LinkedIn content into real outreach. Live. You'll walk away with a repeatable system: what to post, who to reach out to, and how to sequence it. 

Eligible startups also get the LinkedIn-to-Leads Toolkit: ad credits, Apollo, Captions, and HubSpot's Prospecting Agent.

Claim my spot