Futurism says Meta has "almost nothing" to show for AI. The numbers say

Meta's Muse Spark 1.1 scored 51 on Artificial Analysis's Intelligence Index, tying with GPT-5.4, GPT-5.6 Luna, and GLM-5.2, and ranked third on SciCode, contradicting a Futurism article that claimed Meta has 'almost nothing' to show for its AI spending. Futurism's July 31 article dismissed Muse Image as an 'afterthought,' but Meta reported it held No. 2 on Arena's human-preference rankings in three categories at launch. The article also mischaracterized an Axios report, which focused on rival agents' capabilities, not overall model performance.

Futurism wanted a corpse. Meta gave it a messy earnings report, so the publication wrote an obituary. “Despite having almost nothing to show from his enormous spending on AI,” begins the July 31 article https://futurism.com/artificial-intelligence/mark-zuckerberg-pivot-ai-blowing-up-face , headlined “Mark Zuckerberg’s Pivot to AI Is Blowing Up in His Face Spectacularly.” By the end, Meta is described as “practically absent from the frontier AI model race.” Muse Spark is supposedly outdone by competitors in “most tasks.” Muse Image is waved away as an “afterthought.” Meta’s users and engagement are said to “continue to slide.” Almost every part of that picture collapses under inspection. The article does technically mention Meta’s newest models. That is the trick. It gives each one enough space to dismiss it, while withholding the benchmarks, pricing, adoption and distribution that would make the story considerably harder to write. Muse Spark is already a frontier model Muse Spark 1.1 trails the leading models in some long-running agentic tasks. Meta has acknowledged those gaps, and Axios reported that agents from OpenAI, Anthropic and Google can currently operate for longer and handle a wider variety of jobs. Across the whole model, the results look very different from Futurism’s description. Independent evaluator Artificial Analysis scored Muse Spark 1.1 at 51 on its Intelligence Index, effectively tied with GPT-5.4 at xhigh reasoning, GPT-5.6 Luna and GLM-5.2. The model scored 45% on Humanity’s Last Exam, one point behind Claude Opus 4.8. It ranked third among all tested models on SciCode. Muse Spark also carries a one-million-token context window and completed Artificial Analysis’s evaluation with fewer output tokens than the similarly scoring GPT-5.4, GPT-5.6 Luna and GLM-5.2. At Meta’s API prices, Artificial Analysis estimated a cost of roughly 26 cents per Intelligence Index task. That is a credible, economically competitive frontier model with specific weaknesses. Artificial Analysis published the complete results https://artificialanalysis.ai/articles/muse-spark-1-1-everything-you-need-to-know on July 10. Futurism published three weeks later and included none of them. Calling Muse Image an “afterthought” is indefensible Futurism devoted one sentence to Muse Image: “At the same time, Meta released an image-generation model called Muse Image, which felt like an afterthought and a too-little-too-late attempt to catch up with its competitors.” “Felt like” is doing tremendous work there. At launch, Meta reported that Muse Image held the No. 2 position on Arena’s human-preference rankings in three separate categories: text-to-image generation, single-image editing and multi-image editing. Its forthcoming Muse Video model ranked third for text-to-video. Muse Image can search for visual references, write and execute code when a request requires precision, revise its own output, preserve coherence across editing turns and combine people, objects, clothing and environments from multiple reference images. It works jointly with Muse Spark on tasks such as building websites and animated media. Those rankings came through Meta’s announcement and deserved outside scrutiny. A skeptical reporter could have examined the evaluation methodology, tested the product or found evidence that Arena performance does not translate into consumer adoption. Futurism supplied an adjective. The underlying technical announcement https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/ contained enough material to challenge Meta seriously. The article gave readers none of it. The Axios citation does not say what Futurism implies Futurism writes that Muse Spark is “still easily outdone in most tasks by competing models from OpenAI, Anthropic, and Google,” attributing that judgment to Axios. The linked Axios report https://www.axios.com/2026/07/24/meta-muse-spark-agents says rival agents can handle a broader range of tasks and operate autonomously for longer. That is a narrower and more defensible observation about one area of model performance. Another Axios report on the Muse Spark 1.1 launch https://www.axios.com/2026/07/09/meta-ai-spark-model-update-developer noted its improved coding and long-context abilities, competitive API pricing and deployment across Facebook, Instagram, WhatsApp and Meta AI. Axios also identified the strategic advantage missing from Futurism’s story. Meta AI chief Alexandr Wang described the billions of people using Meta’s products, along with the company’s knowledge about those users, as something competitors cannot replicate. Turning a weakness in extended agent operation into “easily outdone in most tasks” stretches the source well past what it says. Meta can put Muse in front of billions of people Meta averaged 3.60 billion daily users across its apps in June, an increase of 3% year over year. Instagram has reached two billion daily users. Facebook also has more than two billion. Threads passed 500 million monthly users. WhatsApp, Instagram, Facebook and Messenger give Meta direct access to a consumer population no standalone AI company can approach. Meta does not have to persuade billions of people to download a new application, create an account and develop a new daily habit. It can place Muse inside products they already open every morning. Distribution does not guarantee affection. Meta AI remains far less culturally prominent than ChatGPT, Claude or Gemini. Plenty of users actively resent its placement inside WhatsApp and Instagram. Meta’s history creates real privacy and trust problems for any product expected to know intimate details about its users. Yet distribution on this scale changes the odds. Leaving it out of an analysis of Meta’s AI position is like evaluating Walmart while forgetting that it has stores. Meta says the number of people interacting with its assistant each day has increased 60% since it rebuilt Meta AI around Muse Spark. More than one million businesses use Meta’s business agents each week. Nine million small businesses use at least one of its AI advertising tools. Those are company-reported figures, and Meta has not disclosed enough raw usage data to establish the depth of consumer adoption. They still belong in any honest account of what the company has produced. The claim that Meta’s engagement “continues to slide” is already stale Futurism linked to an April Verge article about a sequential decline of 20 million daily users during the first quarter. Meta attributed that dip to internet disruptions in Iran and WhatsApp restrictions in Russia. The second-quarter numbers had already been published when Futurism wrote its story. Daily users rose to 3.60 billion. Instagram time spent increased by double digits from the previous year. Facebook video time grew 9% globally and by over 10% in the United States and Canada. Meta’s ad impressions increased 14%. The average price per ad rose 12%. Revenue grew 28% to $60.8 billion, while Family of Apps advertising revenue grew 27%. Whatever one thinks of the quality of Meta’s feeds, “user numbers and engagement continue to slide” conflicts with the latest quarter’s results. The article reached backward for a stale quarter because the current numbers did not cooperate. Meta’s spending still deserves a brutal examination There is a serious bear case here. Meta spent $31.1 billion on capital expenditures during the quarter. Free cash flow fell to $784 million. The company now expects $130 billion to $145 billion in full-year capital expenditures, with enormous additional infrastructure commitments coming behind it. Muse Spark has limited developer adoption compared with OpenAI and Anthropic. Its personal-agent plan is largely aspirational. Meta has not shown that consumers want the “nearly infinite universe” of personalized AI content Zuckerberg described during the earnings call. The prospect sounds horrifying to many of the people already exhausted by synthetic garbage in their feeds. The company also recorded $2.4 billion in legal charges and $1.2 billion in severance expenses. Its workforce is absorbing repeated reorganizations while leadership spends at a scale that could lock Meta into years of depreciation and infrastructure costs. Investors have every reason to demand a credible return. A competitive model and colossal distribution do not automatically justify a $145 billion construction campaign. That is the story Futurism could have written: a profitable company with several genuinely strong models, unmatched distribution and early commercial traction taking an extraordinary financial risk before demand has been established. Instead, it declared that Meta had “almost nothing,” ignored inconvenient performance data, misstated fresh engagement figures and converted a specific weakness in agentic endurance into failure across “most tasks.” Meta may eventually burn a historic amount of money on an AI strategy that never earns an acceptable return. That outcome remains possible. Declaring the strategy a spectacular failure weeks after Meta shipped SOTA-class models, while its users, engagement and revenue are growing, is premature. It is a mood packaged as analysis. Reddit rewarded the mood. Readers deserved the ledger.