What AI X was talking about, Sep 21–28

Eight stories, ranked by how much X was talking: post volume and the reach of the top posts.

Numbers from the X API, pulled Mon Sep 28, 2026. Times CST.

Story 1 of 8

Meta Muse: press vs X

Press says

Muse is a hit: #1 free iPhone app with over 2.5M downloads, and Meta stock had its best month since 2013. Reviewers found it good at planning-style tasks.

On the Marketplace address incident, coverage stayed balanced: Meta is investigating, and Muse may have followed permissions.

X says

The biggest Muse posts of the week were negative.

Backlash

  • Muse sold a seller's keyboard for a lowball price, gave the buyer his street address, and auto-replied “Yep I'm here!” when he wasn't home. “Dangerous and creepy.” Some said they deleted the app.
  • 404 Media: some “AI” phone calls were really handled by humans in call centers.
  • Patrick Wardle posted Mac/iOS exploit chains. Early users called it “underwhelming and useless.”

Defenders

  • The phone-call feature is “ridiculous” in a good way; Marketplace haggling stories (a PS5 for about €50 less). Mac computer use landed well.
  • Meta's David Singleton: past investigations “consistently” found Muse followed direct instructions.
About 79,957 posts this week · X API count, Sep 21–28
Story 1 of 8 · top posts

Muse on X: backlash

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 1 of 8 · top posts

Muse on X: defenders

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 2 of 8

Claude Opus 5.5: press vs X

Press says

Opus 5.5 is 20% cheaper per token, about 40% cheaper per task, and matches Fable 5.1 on most work.

The New Stack flagged that safety classifiers can quietly reroute some requests to older models.

X says

Less about price. More that Opus 5.5 has “taste,” Claude limits feel “unlimited,” and power users are moving from Codex back to Claude Code.

Pro

  • DHH: “Anthropic did magic.”
  • Theo: Claude went from less popular than Codex in T3 Code to 2x more popular in two weeks.
  • levelsio: “Claude Code is now cool again.”

Skeptics

  • Opus 5.5 is “talkative & annoyingly needy.”
  • Theo's jab at every lab: “Anthropic has no small models that are worth using right now.”
  • The classifier fallback to a less capable model stayed niche on X.
About 106,621 posts mentioning “Opus 5.5”, the most of any story checked · X API count, Sep 21–28
Story 2 of 8 · top posts

Opus 5.5 on X: pro

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 2 of 8 · top posts

Opus 5.5 on X: skeptics and a data point

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 3 of 8

Opus 5.5 motion graphics: press vs X

Press says

Mainstream outlets checked didn't cover this. Launch stories stuck to price and benchmarks.

The closest coverage is prompt-library and how-to posts explaining that Claude builds these by writing code for render tools, not by generating video natively.

X says

Creators kept pasting the same one-sentence prompt and posting the results:

“make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. go all out.”

“Cooked” camp

  • Motion design as a paid skill is in trouble, and OpenAI “has awful taste” in side-by-side comparisons.

Skeptics

  • They want to see the real process behind “one-shot” claims; the output isn't flexible or editable. One self-declared skeptic (“liars!!”) tried it and posted his result.
About 2,055 posts; 73 on Sep 24 (UTC), 431 on Sep 25 · X API count, Sep 21–28
Story 3 of 8 · top posts

Motion graphics on X

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 4 of 8

GPT-6 Sol: press vs X

Press says

OpenAI cut Sol and Luna prices in half and said Sol “makes about half as many mistakes as its predecessor, reaching Astra-level reliability.”

X says

Power users called it a regression.

Complaints

  • Bug-fix tests where Sol scored below GPT-5.6 Sol and Opus 5.5.
  • Around #60 in Text Arena vs about #19 for GPT-5.6 Sol.
  • It “overthinks,” “feels lazy,” and needs babysitting.

Defenders

  • Mostly LMArena's own numbers (#6 in Agent Arena, up over 5.6 Sol), plus “underrated” and “a price cut, not a leap.”

Spillover

  • Many “switching back to Claude Code” posts name Sol.
About 26,168 posts; peak about 8,400 a day on Sep 22–23 (UTC) · X API count, Sep 21–28
Story 4 of 8 · top posts

GPT-6 Sol on X

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 5 of 8

OpenAI DevDay leaks: press vs X

Press says

Mostly leak-tracking: a $500/month tier, reportedly with “Fastest Work and Codex,” possibly on Cerebras. Not confirmed by OpenAI.

New sign-ups for the $200 Pro plan have been paused since Sep 10.

DevDay is Tue Sep 29 at 10:00 PT (11:00 CST).

X says

The loudest take: OpenAI is behind on consumer agents.

Skeptics

  • Muse and Grok Bot shipped always-on agents and OpenAI hasn't, despite hiring OpenClaw creator Peter Steinberger. The “o” leak got framed as OpenAI's answer.
  • The $500 price drew “what is speed worth” analysis, and anger that OpenAI “quietly removed 5x and 20x” from its pricing page.

OpenAI side

  • “We are not over. We are pacing.” (@xiangyuqi_pton, described in X's news summary as an OpenAI researcher). OpenAI Devs posted teasers.
About 4,312 posts matching Pro Max, Aeon + OpenAI, or DevDay · X API count, Sep 21–28
Story 5 of 8 · top posts

DevDay leaks on X

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 5 of 8 · top posts

“It's over for OpenAI” and the replies

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 6 of 8

OpenSEO joins YC: press vs X

Press says

No tech-press coverage of the YC news found.

What exists is the project's own framing and roundup posts: an open-source, self-hosted Semrush/Ahrefs alternative for $10/month.

X says

X argued whether a “DataForSEO wrapper” deserves YC.

Fans

  • Indie hackers and SEO users like a $10, MIT-licensed tool, and 3,000+ paying users is real traction.

Wrapper camp

  • It's “a UI on top of DataForSEO,” the hosted version marks up calls 28%, and “anyone with claude can build this in a few days.”
  • One founder vibe-coded a clone in a few hours. DataForSEO's data quality gets questioned next to Ahrefs.

Founder

  • Ben Senescu: “PSA: Build a wrapper.”
About 400 posts (peak 237 on Sep 24 UTC); biggest critical post about 10K views · X API count, Sep 21–28
Story 6 of 8 · top posts

OpenSEO on X: the numbers

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 6 of 8 · top posts

OpenSEO on X: the wrapper argument

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 7 of 8

Space Bunny: press vs X

Press says

Only niche blogs covered it: a free stealth model with 1M context whose fingerprints point to MiniMax (M3.1). Unconfirmed.

It showed up on OpenCode/OpenRouter around Wed Sep 23.

X says

A free mystery model that builds games in one HTML file.

Hype

  • Builders ran side-by-sides saying it “cooked” GPT-6 Astra on Three.js scenes, for free.

Knocks

  • Physics sims were “TERRIBLE,” it lost a Minecraft test to DeepSeek V4.1 Flash, and it can be slow (45 to 55 minutes for one build).

Who made it

  • When MiniMax M3.1 Flash actually appeared, testers said the two didn't match.
About 2,346 posts this week · X API count, Sep 21–28
Story 7 of 8 · top posts

Space Bunny on X

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Story 8 of 8

ChatGPT Voice plugins: press vs X

Press says

Voice can now use email, calendar, and Slack plugins, runs on GPT-6 Astra, Sol, and Luna, and works inside ChatGPT Work.

High-risk actions still need on-screen approval.

X says

Mostly positive, from people who used it for hours. Small on X next to the others.

Praise

  • Viticci: it “works with any plugin I've tested,” including his custom Mac MCP.
  • Others said it's underrated in the OpenAI-vs-Anthropic debate.

Dissent

  • Grok Bot's new voice is more agentic (it can run a full VM and MCPs); some report memory drops and looping.
About 1,072 posts this week · X API count, Sep 21–28
Story 8 of 8 · top posts

ChatGPT Voice on X

Views and likes as of Sep 28, 2026 · times CST · tap a post to open it on x.com
Also this week

Runners-up: checked, smaller on X

AI on X · Sep 21–28