Artificial Analysis said on 28 October 2026 that Meta’s Muse models are “the most Friendly of all models” in its AA-PersonalityBench, with Muse Spark 1.3 at 100, Muse Spark Open at 98, Muse Wonder at 95 and Muse Glimmer at 93.
On the chart Artificial Analysis posted, OpenAI’s GPT-6 Astra scores 58 on the same trait. By our arithmetic, Muse Spark 1.3 sits 42 percentile points above it.
A vertical news Short narrated from the published headlines and opening lines of this piece and three related FourWeekMBA pieces.
Business Pill · Distribution Advantage
An evergreen Business Engineer explainer on distribution advantage: why reach decides who gets tried. It explains the general idea only and does not discuss Meta, Muse or this ranking.
The key insight: By our arithmetic, Muse Spark 1.3 scores 42 percentile points above GPT-6 Astra on Artificial Analysis’ Friendly trait, while Similarweb puts US Muse users at 7.5 minutes a day, about a third of ChatGPT‘s 23.
Artificial Analysis says the Friendly trait “measures whether a response conveys warmth and approachability without feeling performative”. Scores are percentiles from 1 to 100 across the models in the benchmark; its page shows 41 models.
The benchmark page says AA-PersonalityBench “is not a capability benchmark: scores describe how a model tends to write and respond, not whether it is correct”. It says a panel of LLM judges compares two answers to the same prompt and picks the one that shows more of the trait, “whether or not the trait is desirable”.
At the other end, Artificial Analysis lists Nemotron 3 Ultra at 1, GPT-3.5 Turbo at 3, MiniMax-M3.1-Flash at 6, Grok 4.6 at 8 and DeepSeek V4.1 Flash 1017 at 11 as the least Friendly.
AA-PersonalityBench Friendly trait, percentile 1-100, highest-scoring model per lab on the chart Artificial Analysis posted on 28 October 2026. Our arithmetic: Muse Spark 1.3 at 100 is 42 percentile points above GPT-6 Astra at 58. The benchmark measures writing style, not accuracy or usage. Source: Artificial Analysis.
What Was Posted on X
muse models are the most friendly of all the models!
Alexandr Wang posted on 29 October that Muse models are “the most friendly of all the models!” and added “personal superintelligence”.
So much for $Metaโs โdistribution advantage.โ
Even at its launch peak, Muse never managed to exceed ChatGPTโs daily search interest in the US.
Today, the gap is not even close.
Muse had a sharp launch spike, then faded quickly from that peak. That does not automatically mean the product is failing – reacceleration is still possible.
But when interest spikes this hard and then rolls over this quickly, it is usually not the cleanest sign of durable consumer pull.
The account @ritmus_sk wrote on 28 October that, even at its launch peak, Muse “never managed to exceed ChatGPT’s daily search interest in the US”. The post says Muse’s spike faded quickly, which it calls “usually not the cleanest sign of durable consumer pull”, while noting that “reacceleration is still possible”.
How Muse Compares With Rival Flagships
On the chart, the highest-scoring model outside Meta is GLM-5.3 at 83. Anthropic’s Claude Opus 5.5 scores 73, Google’s Gemini 4 Argon 70, GPT-6 Astra 58, Kimi K3.1 Pro 21 and xAI’s Grok 4.7 13.
By our arithmetic, Meta’s four Muse models take the top four places, and the lowest of them, Muse Glimmer at 93, is 10 points above GLM-5.3.
What Was Posted on X
Alexandr Wang, whom the Artificial Analysis post congratulated, quoted it on 29 October and wrote: “muse models are the most friendly of all the models!”, adding “personal superintelligence”.
A day earlier, the X account @ritmus_sk posted a Google Trends chart of US search interest and wrote: “Even at its launch peak, Muse never managed to exceed ChatGPT’s daily search interest in the US”. The post added: “That does not automatically mean the product is failing – reacceleration is still possible”.
Search Interest and Daily Use
The chart in that post covers August to October 2026. Its subtitle says it shows “a sharp spike for” Muse “around its launch on September 8”, while ChatGPT “remains at a much higher level”.
Other measures we have reported point in different directions. Sensor Tower data reported by TechCrunch put Muse at 10.2 million US downloads since its September debut; ChatGPT had 3.7 million US downloads in its first 40 days. The same data put Muse’s 7-day retention at 30% against a 19% rival average, as we reported on 20 October.
Similarweb’s 27 October breakdown put US users of the Muse app at 7.5 minutes a day, against 23 minutes for ChatGPT, as we reported on 27 October. By our arithmetic, that is about a third of ChatGPT’s daily time.
What Meta Said
In its third-quarter 2026 results release, Meta quoted chief executive Mark Zuckerberg: “We hit major milestones this quarter toward delivering personal superintelligence for everyone – including launching Muse and Muse Wonder, our first frontier model”.
Sensor Tower data reported by TechCrunch showed Meta put 39% of its total advertising spend behind Muse in October.
As we read the sources together, Muse leads on the Friendly trait, per Artificial Analysis, and had more US downloads since its September debut than ChatGPT had in its first 40 days, per Sensor Tower, while Similarweb’s daily minutes and the @ritmus_sk search chart show ChatGPT ahead today.
As we read Artificial Analysis’ method, the Friendly score records which of two answers a panel of LLM judges sees as showing more of the trait; the page says it is not a measure of correctness.
As we read Meta’s release, Zuckerberg ties Muse to “personal superintelligence”, the same phrase Wang used in his post about the ranking.
Mark Zuckerberg, Meta founder and CEO, in Meta’s third-quarter 2026 results release
“We hit major milestones this quarter toward delivering personal superintelligence for everyone – including launching Muse and Muse Wonder, our first frontier model”
Three Implications
FOR META Sensor Tower data reported by TechCrunch showed Meta put 39% of its total advertising spend behind Muse in October, with Muse’s 7-day retention at 30% against a 19% rival average.
FOR RIVAL LABS On the Artificial Analysis chart, Anthropic’s Claude Opus 5.5 scores 73, Google’s Gemini 4 Argon 70 and OpenAI’s GPT-6 Astra 58 on the same trait.
FOR BENCHMARK READERS Artificial Analysis says AA-PersonalityBench “is not a capability benchmark”, so a Friendly score sits beside, not in place of, accuracy and usage figures.
The framework describes two curves, what the underlying technology can do and what users can actually do with it, and names the gap between them: “what the platform makes possible versus what the product surfaces”.
As we read it, a Friendly score describes the model, while daily minutes and search interest describe how people use the product. The framework was published in June 2026 and does not discuss Meta or Muse.
What Is Not Established
AA-PersonalityBench scores how models write, not whether answers are correct or how many people use them. The scores are percentiles across the models in the benchmark, so a model’s score can change as models are added.
The Google Trends chart was made and posted by @ritmus_sk; we did not reproduce it. Search interest is an index, not a count of users, and the word Muse is also searched for reasons unrelated to Meta’s app. Sensor Tower and Similarweb are third-party estimates with different methods and dates.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Artificial Analysis ranks Meta’s Muse models as the four most Friendly models in AA-PersonalityBench. By our arithmetic, Muse Spark 1.3 scores 42 percentile points above GPT-6 Astra, while Similarweb puts US Muse users at about a third of ChatGPT’s daily time.
A note on sourcing. Friendly percentiles and the benchmark description come from Artificial Analysis’ 28 October 2026 post, the chart attached to it and its AA-PersonalityBench page. The search-interest chart is @ritmus_sk’s, posted on X; we did not reproduce it. Downloads, retention and advertising share are Sensor Tower data as reported by TechCrunch, and daily minutes are Similarweb’s, both as covered in our earlier pieces. Point gaps and the time ratio are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
