The $417 Billion Contradiction
OpenAI launches GPT-6.1 Terra at half the price of Sol
OpenAI on Monday, October 26, released GPT-6.1 Terra, a mid-tier model built for computer use. The company says the model "understands what's on screen, takes action across apps, writes naturally," and costs less to run than the models above it. OpenAI is positioning Terra for high-volume office work, including drafting proposals, interpreting technical drawings, filling out forms and editing videos. Its standard API price is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol.
Pricing and Availability
The API price is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. GPT-6.1 Sol costs $2 for input and $10 for output. Developers can use the model as gpt-6.1-terra through the Responses and Chat Completions endpoints. Independent tracker Artificial Analysis lists the model with a 1 million-token context window. It notes that Terra's prices sit below the median of $2 for input and $10 for output among the models it tracks.
In ChatGPT, Terra is rolling out in Work and Codex to Plus, Pro, Business, Enterprise and Edu plans. It is not offered in the standard Chat experience. Overall usage limits stay the same, but OpenAI says each allowance covers more tasks with Terra.
Benchmark Claims
On Agents' Last Exam, a benchmark that tests how well agents use software at work, OpenAI says Terra matches its flagship GPT-6 Astra at one-tenth the price per token. It also finishes tasks more than three times faster on average than Astra and GPT-6.1 Sol. On OpenAI's internal test of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%. On Terminal-Bench 4.0, Terra solved 50.3% of tasks at max effort, compared with 21.5% for its predecessor. OpenAI still recommends Sol and Astra for the most complex coding work.
Artificial Analysis scored Terra at 48 on its Coding Agent Index. That puts it 6 points behind GPT-6.1 Sol and 17 ahead of GPT-6 Luna.
Safety and Context
Terra is the first upgrade to OpenAI's Terra tier since GPT-5.6. It arrives about four weeks after GPT-6.1 Sol. OpenAI's lineup now runs from Luna through Terra and Sol to Astra.
OpenAI's system card addendum says the company is treating Terra as "Critical" in cybersecurity and "High" in biological and chemical capability under its Preparedness Framework. Because of that rating, the model uses the same safeguards as GPT-6.1 Sol and GPT-6 Astra. The launch came on the same day that rival Anthropic filed its S-1 ahead of a planned initial public offering.
OpenAI Says AI Has Broken Mathematics, Computer Science, and Physics. The Real Test Is What Happens Next.
Three alleged breakthroughs. Three foundational scientific problems. One uncomfortable question: when AI claims to have solved problems that generations of brilliant researchers could not, who decides whether the revolution is real?
OpenAI’s October 27 announcement is extraordinary even by the standards of AI hype. Its models have reportedly produced a proof of the local smoothing conjecture in every dimension, refuted the Exponential Time Hypothesis, and established the mathematical existence of quantum field theories encompassing quantum chromodynamics and a unified completion of the Standard Model.
If these results survive independent scrutiny, they would represent a genuine inflection point in the history of scientific discovery. But the most important story here is not simply what AI claims to have proved. It is how science must change when machines begin producing results that human experts need time, resources, and specialized collaboration to understand.
The Claims Are Enormous. So Is the Verification Burden.
Consider the computer science result. The Exponential Time Hypothesis, or ETH, is a foundational conjecture in computational complexity. Researchers have relied on it to establish conditional hardness results across fields ranging from graph theory to computational biology. OpenAI says its system has refuted the hypothesis and formalized the result in Lean, a proof assistant that can check mathematical reasoning against a formal logical framework.
That is potentially seismic. But a machine-checked proof is not the same thing as a field-wide understanding of what has been established, which assumptions matter, or what consequences follow. Formal verification can provide powerful evidence that a proof follows from its encoded premises. It does not eliminate the need to scrutinize the formalization, interpret the result, and establish its significance.
OpenAI has assembled a group led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams to examine the ETH result. That is a sensible move. The people best positioned to assess a claim of this magnitude are researchers who understand the history, technical machinery, and implications of the problem.
The mathematics claim is equally ambitious. OpenAI says its models proved the local smoothing conjecture in all dimensions, thereby resolving the Kakeya and Stein restriction conjectures in all dimensions as well. The company says it has released both a natural-language proof and a Lean formalization.
The physics claim may be even more demanding to assess. OpenAI reports a rigorous existence proof for interacting quantum field theories involving gauge fields and fermions in four dimensions, including QCD and a unified completion of the Standard Model. The work also includes pure Yang–Mills theory, part of a famous Millennium Prize problem.
These are not three ordinary research papers with unusually impressive abstracts. They are claims whose implications could reshape entire research programs. That makes independent examination essential, not optional.
The Most Interesting Part Is OpenAI’s New Research Model
Buried beneath the headline-grabbing claims is a more consequential institutional experiment.
OpenAI is testing three different approaches to releasing AI-generated scientific discoveries. For mathematics, it is publishing the AI output and leaving researchers to organize their own efforts to explain it. For computer science, it has established an external collaboration to interpret the result while making the raw output public. For physics, it is deferring to an expert-led group that has chosen to prepare its own write-up before the AI-produced artifacts are released.
This is an implicit acknowledgment of a problem that conventional scientific publishing was not designed to solve.
Scientific papers normally emerge from a process in which researchers develop an argument, explain its context, defend its assumptions, and communicate its significance. Here, the discovery process may run ahead of the explanation process. AI can produce a purported result faster than a community can fully interpret it.
The bottleneck shifts from discovery to understanding.
That distinction matters. A proof that exists but has not been understood by the relevant community is not worthless. But neither is it equivalent to a result whose validity, scope, and consequences have been thoroughly established. Scientific knowledge requires both technical correctness and a shared capacity to inspect, reproduce, criticize, and build upon the work.
OpenAI appears to recognize this. The company says the collaborations will operate transparently, without nondisclosure agreements, and that participants can discuss their work and publish their conclusions. It also says it does not seek academic credit for discoveries produced by its internal models, while researchers should receive recognition for the explanations and materials they create.
Those commitments are promising. They are also commitments whose value will be determined by how they work in practice.
The Credit Problem Is Bigger Than It Looks
Who deserves credit when an AI system produces a discovery, a formal proof checker validates a formalization, and human researchers spend months turning the result into something the field can understand?
OpenAI says it does not seek academic credit for the discoveries generated by its models. That is a meaningful distinction, but it does not settle the deeper question of scientific attribution.
Credit is not merely a matter of putting names on a paper. It determines who gets hired, promoted, funded, invited to speak, and remembered in the history of a discipline. It also helps establish who is accountable when an argument fails.
The right framework must distinguish between generating a result, formalizing it, checking it, interpreting it, explaining it, and extending it. Those contributions may come from different people and systems. They should not be collapsed into a single authorship label.
OpenAI’s proposed approach at least acknowledges that distinction. Whether the broader scientific ecosystem can develop fair and consistent conventions around it remains an open question.
There is another issue: the selection of initial experts inevitably shapes the process. OpenAI itself acknowledges that choosing initial shepherds creates asymmetries of information and influences who gets involved first. Transparency helps, but it cannot by itself guarantee broad participation or equal access to the most consequential work.
If AI-generated discoveries become a major source of scientific progress, access to the artifacts, verification infrastructure, computing resources, and opportunities to contribute will become questions of scientific governance.
The Physics Result Demands Particular Care
The reported quantum field theory result deserves special attention because the distance between a mathematical existence theorem and a complete physical understanding can be enormous.
Constructive quantum field theory addresses deep questions about whether theories used throughout physics can be defined rigorously under appropriate mathematical conditions. Yang–Mills theory is particularly significant because its mathematical foundations are connected to one of the Clay Mathematics Institute’s Millennium Prize problems.
OpenAI says its models produced a natural-language proof for Yang–Mills existence in September, followed by a QCD proof, and completed a Lean formalization of Yang–Mills existence in October. For this release, an expert group led by David Gross will take responsibility for understanding and explaining the work. The group has chosen to delay publication of the AI-produced artifacts while preparing its own account.
That is a different release strategy from publishing the mathematical and computer science outputs immediately. It may be appropriate for a result whose technical complexity makes premature interpretation especially risky.
But the key question remains the same: what exactly has been proved, under which assumptions, and how does the result relate to the precise mathematical statements at stake? The announcement describes the scope of the achievement. The scientific community must establish its meaning and validity in detail.
The distinction between a headline and a theorem is where the real work begins.
Funding the Verification Is Part of the Breakthrough
OpenAI also points to a recent five-million-dollar contribution to the Simons Foundation, intended to support workshops, conferences, and programs that help researchers understand and build on AI-generated results. For the ETH work, the company says it has agreed to fund a conference devoted to understanding the implications of the claim.
This is important because independent verification is not free. Experts need time. Collaborations need coordination. Formal proof artifacts require scrutiny. Results with broad implications may require workshops, explanatory papers, and years of follow-up research.
If AI systems can generate scientific claims at a pace that outstrips the research community’s ability to evaluate them, funding verification becomes part of the infrastructure of discovery itself.
But funding must not become a substitute for independence. The credibility of this process depends on researchers being able to reach conclusions that are genuinely their own, including conclusions that challenge, narrow, or reject the original claims.
OpenAI says the external collaborators participate voluntarily and without pay, and that their conclusions are independent. The company also says the collaborators will receive recognition for the work they produce. Those are useful principles. The test will be whether the resulting process gives outside experts the freedom, access, and time necessary to reach an honest verdict.
What Would Actually Change If These Results Hold?
If the claims withstand independent examination, the implications extend far beyond three famous conjectures and theories.
Mathematics could see a new division of labor between discovering proofs and making them intelligible. Theoretical computer science could be forced to revisit results whose hardness guarantees depend on ETH. Mathematical physics could gain new tools for establishing rigorous foundations for theories that have long been central to our understanding of nature.
More broadly, AI could become a system for generating candidate scientific knowledge rather than merely assisting with calculations, literature reviews, or routine derivations.
But none of this means every AI-generated breakthrough will be correct, useful, or transformative. The more powerful these systems become, the more important it is to distinguish between a result that sounds revolutionary and one that survives adversarial scrutiny.
The appropriate response is neither reflexive disbelief nor breathless acceptance. It is organized skepticism: public artifacts, independent experts, transparent methods, reproducible checks, explicit assumptions, and enough time for criticism.
The Announcement Is the Beginning, Not the Conclusion
OpenAI’s announcement could mark an important shift in the relationship between AI and scientific research. Its most consequential contribution may turn out to be not any single theorem, but the model it is testing for turning machine-generated discoveries into durable scientific knowledge.
The company is effectively acknowledging that producing a result and establishing its place in science are different jobs. That is the right distinction to make.
Now comes the hard part: independent verification, technical interpretation, reproducibility, and a fair system for deciding who gets access and credit.
Until those processes run their course, the claims should be treated as major reported breakthroughs awaiting full scientific assessment, not as settled facts simply because the announcement is extraordinary.
The real measure of AI’s scientific power will not be how many historic problems it claims to solve. It will be how many results the scientific community can independently verify, understand, and use to discover things that were previously out of reach.
That is the standard these claims must meet. And if they meet it, the consequences could be profound.
Apple's first foldable cracked after 6 hours, one owner says.
The phone is the iPhone Duo, and it went on sale on October 23. A buyer in Australia told MacRumors he had opened it fewer than 20 times when he heard a crack and the screen broke right along the hinge. Apple swapped it on the spot. Other owners have posted the same kind of crack on Reddit, and people are calling it "Crackgate".
The Verge reports that a YouTuber called CuriousHagi got the same crack while testing one, then took the phone apart. The hinge is built like a chain. One link in the middle is thinner than the others, and it snapped at a small notch. It's like a bike chain where one link is made of thinner metal.
The catch is that this is one YouTuber's theory. PCMag says Apple hasn't commented or said how many phones are affected, and Gulf News says some viral crack photos are fake. The sources are in the first comment.
𝐓𝐡𝐞 𝐆𝐫𝐨𝐰𝐭𝐡 𝐍𝐮𝐦𝐛𝐞𝐫𝐬 𝐀𝐫𝐞 𝐒𝐭𝐮𝐧𝐧𝐢𝐧𝐠
Anthropic’s revenue went from $41 million in 2023 to $386 million in 2024 and $4.59 billion in 2025. In the first six months of 2026 alone, it reported $16.29 billion, roughly fourteen times the same period a year earlier. Few companies in history have scaled this fast, and the people celebrating the IPO filing have good reason to. A majority of that money comes from businesses paying per token for model access, which is the kind of usage-based revenue that compounds when customers build products on top of you.
But a fast-rising line on a chart is not the same as a durable business, and the filing invites us to ask harder questions than the headline numbers do. Revenue grew fourteen-fold while expenses grew fivefold. That is a remarkable operating story, but it also tells you how much of the company’s future depends on sustaining that gap while the cost base keeps climbing.
𝐓𝐡𝐞 𝐂𝐨𝐦𝐦𝐢𝐭𝐦𝐞𝐧𝐭𝐬 𝐀𝐫𝐞 𝐁𝐢𝐠𝐠𝐞𝐫 𝐓𝐡𝐚𝐧 𝐭𝐡𝐞 𝐁𝐮𝐬𝐢𝐧𝐞𝐬𝐬
Anthropic says it has committed more than $600 billion in spending with about eight partners over the next decade. That includes at least $110 billion with Amazon, roughly $114.5 billion with Google, and $44.9 billion with Microsoft for cloud computing. Many of these agreements are non-cancelable or require payment regardless of how much the company actually uses.
Do the rough arithmetic. Spread $600 billion over ten years and you get about $60 billion a year. The company’s first-half 2026 revenue, annualized, is about $33 billion. Revenue will need to grow substantially just to cover the average annual commitment, before any profit. That is possible if the growth curve holds, but it means the company is making fixed promises today based on revenue projections it has not yet proven it can sustain. If usage slows, pricing pressure arrives, or a competitor takes share, those non-cancelable payments do not shrink to match.
𝐓𝐡𝐞 𝐂𝐢𝐫𝐜𝐥𝐞 𝐏𝐫𝐨𝐛𝐥𝐞𝐦
Two of the company’s largest cloud suppliers are also among its largest backers. Amazon has invested at least $13 billion and agreed to invest up to $20 billion more. Google has committed up to $40 billion. Money flows from these companies into Anthropic, and then back out to them as cloud fees. Revenue at the supplier looks like demand, and investment at the buyer looks like confidence, but part of the picture is the same capital moving in a loop.
This pattern is not unique to Anthropic. It runs through much of the AI industry, and it makes the growth numbers harder to read cleanly. When a supplier is also an investor, it has a strong reason to keep the customer spending. Investors should ask how much of the demand signal is organic and how much is structural.
The market-share figure deserves the same skepticism. Ramp data showed Anthropic accounted for 60.1% of AI token spending by business customers in the week ended October 11. That is a striking number, but it comes from one data source that captures only a slice of spending, measured over a single week. Concentration of that kind can be a moat or a dependency, and a single week does not tell you which.
𝐏𝐮𝐛𝐥𝐢𝐜 𝐈𝐧𝐯𝐞𝐬𝐭𝐨𝐫𝐬 𝐆𝐞𝐭 𝐚 𝐒𝐞𝐚𝐭 𝐚𝐭 𝐭𝐡𝐞 𝐄𝐝𝐠𝐞
Anthropic is a public benefit corporation, which means its board is legally required to balance profit with developing AI for humanity’s benefit. That is an admirable commitment in principle. In practice, the filing describes a structure in which a majority vote of seven co-founders directs a single Class F share carrying 50.1% of the total voting power over key corporate matters. That includes electing some board directors.
The company argues that concentrating control in fewer hands reduces AI’s risks. That may be true. It is also true that the arrangement limits the influence public shareholders can have over strategy, spending, and leadership. Buyers of the stock would be paying for a share of a company whose most important decisions are not theirs to make. That is a legitimate trade for some investors, but it should be named plainly rather than buried under the language of mission.
Anthropic is one of the most impressive commercial stories in the current AI cycle. Its revenue growth is real, its enterprise position is strong, and its leadership has a clear thesis about where the industry is going. None of that is in question here.
The question is whether a company making more than $600 billion in non-cancelable commitments, funded partly by the same firms it buys from, governed by a small group of founders, and priced on a growth curve that has not yet had to survive a downturn, is being valued for what it has proven or for what it hopes to become. The filing gives investors the numbers to answer that for themselves. The smart move is to read the commitments section as carefully as the revenue chart.
Musk wants Tesla-SpaceX deal before Trump exits: Bloomberg
Elon Musk has given his public support to combining Tesla and SpaceX. He said both companies have set up special committees of independent directors to look at a deal, and he cautioned that a merger is far from certain. "Do I think it would be a good idea? Yes," Musk said on SpaceX's third-quarter earnings call on Tuesday, October 20, adding that "the special committees have not reached agreement on terms of any transaction and there's no guarantee that they will".
According to Reuters, via Virginia Business, this was the first official confirmation that the companies were in formal talks. A person familiar with the matter told Reuters the discussions began in recent weeks and could lead to SpaceX acquiring Tesla.
Why Musk Supports It
Musk runs both companies. He said the committees were formed at shareholders' request and that he is excluded from their talks because he is a "conflicted party". He said there is "a lot of merit to combining the companies as we are pretty much in the superintelligence singularity." He pointed to possible cooperation on AI software, chip design, robotics and connectivity.
Bloomberg reported on Monday, October 26, that Musk wants to close the deal before President Donald Trump's term ends. Citing people familiar with the matter, Bloomberg said an announcement is being targeted for the first half of 2027.
Wall Street Turns Bullish
Tesla shares rose about 5% on Monday and hit their highest level in more than three months after Barclays upgraded the stock to overweight. Barclays raised its price target to $450 from $370 and called the expected merger "a key part of the Tesla bull case". Bloomberg found that the share of sell ratings on Tesla has dropped to its lowest in more than 11 years. Last week, BNP Paribas and CLSA dropped their bearish calls. Dan Ives of Yorkville Ives puts the chance of a merger by the end of 2027 at 80% to 90%.
Votes and Regulators
On the Tesla side, a deal needs approval from a majority of all outstanding shares, not just the shares that are voted. Musk called that "quite a high hurdle." He controls about 85% of SpaceX's voting power but only about 20% at Tesla. JPMorgan analyst Rajat Gupta sees approval from China as the biggest obstacle. He pointed to Tesla's Shanghai factory, SpaceX's U.S. defense ties, and the fact that Starlink has never been approved to operate in China.
Reuters analysis found that Musk's ties to Trump may ease U.S. regulatory concerns. However, competition authorities in China and South Korea could look closely at whether Tesla would gain an unfair advantage from access to SpaceX technology. Analysts at William Blair argued that the push is "all about control".
INSTAGRAM WANTS TO BE YOUTUBE. THE CREATOR ECONOMY SHOULD BE TERRIFIED.
Instagram is testing 90-minute videos. The headlines will call it a war on YouTube. The business reality is more interesting, and considerably more cynical: Meta wants to own the entire creator economy, from the first three-second hook to the hour-long video, the fan relationship, the advertising inventory, and ultimately the money.
This is not a revolution in video. It is a land grab for attention.
THE GREAT CREATOR ECONOMY LIE
For years, Instagram has trained creators to produce an endless supply of free entertainment. Shoot a Reel. Chase the algorithm. Manufacture a viral moment. Repeat tomorrow. The platform gets engagement, advertisers get audiences, and creators get the intoxicating possibility of going viral.
Possibility is not compensation.
Instagram chief Adam Mosseri has acknowledged the uncomfortable economics: the platform’s performance-based creator payments remain limited, and scaling them would mean spending substantially more than Instagram earns back. Meanwhile, YouTube has built a much clearer proposition around advertising revenue, including a 55% share of eligible long-form video ad revenue for creators in its Partner Program.
The difference is fundamental. YouTube has spent years building a business in which a creator’s library of videos can become a recurring revenue-generating asset. Instagram has largely perfected the business of making creators visible.
Visibility is valuable. But visibility without predictable monetization is a particularly elegant way to extract labor from people who desperately want an audience.
NINETY MINUTES WON’T FIX A BROKEN PAYCHECK
Here is the question Instagram would prefer everyone to postpone: Who gets paid, how much, and according to what formula?
A 90-minute upload limit is a product feature. A sustainable creator income model is a business commitment. Confusing the two is how platforms generate breathless headlines without fundamentally changing the economics for the people making the content.
Meta says it is exploring ways creators could earn money from brands, fans, and Instagram itself. That is an acknowledgment of possibility, not a monetization policy.
Will creators receive a meaningful share of advertising revenue? Will payments depend on watch time, qualified views, subscriptions, or opaque performance criteria? Will smaller creators have a realistic path to earning money, or will the rewards concentrate among accounts that already command enormous audiences?
Until those questions have concrete answers, the 90-minute video is not a creator-economy breakthrough. It is an invitation to produce more content for a platform that has yet to promise what that extra labor will be worth.
THE REAL WEAPON ISN’T VIDEO. IT’S DISTRIBUTION.
Instagram’s most strategically interesting move is not the long-form upload itself. It is the proposed relationship between long videos and Reels.
The platform is testing a feature called Clips as Reels, allowing creators to turn segments of longer videos into short-form content that links back to the full version. Instagram is also exploring ways to connect episodic content and improve video discovery.
This is smart. Potentially dangerous for competitors, too.
A creator can use short clips to attract attention, then move viewers into a longer viewing session without sending them to another platform. Instagram gets more reasons for users to stay. Advertisers get more opportunities to reach them. Meta gathers more signals about their interests and behavior. Creators get a potentially more coherent funnel from discovery to fandom.
But notice who benefits immediately: the platform.
Creators benefit only if the additional attention translates into income, audience ownership, or commercially useful relationships. Otherwise, Instagram has simply become a more efficient machine for keeping creators and viewers inside its walls.
The winning platform will not necessarily be the one that hosts the longest videos. It will be the one that converts attention into money most reliably.
YOUTUBE HAS A MOAT. IT ALSO HAS A PROBLEM.
YouTube’s advantage is not merely that it allows long videos. It has a mature advertising infrastructure, established creator monetization, search-driven discovery, and a deep archive of content that can continue attracting viewers long after publication.
A successful video can keep earning. A tutorial can answer a question for years. A documentary can accumulate views long after the creator has moved on to the next project.
That is an entirely different economic proposition from betting that tomorrow’s Reel will outperform today’s.
Instagram can challenge this advantage, but it cannot simply copy the runtime. It must replicate the economic machinery underneath it.
And YouTube should not feel invincible. Instagram already has something YouTube cannot manufacture overnight: a social graph built around relationships, messaging, personalities, and habitual daily engagement. If Meta can combine that distribution advantage with compelling long-form viewing and competitive creator payouts, it could make YouTube less indispensable.
The threat is real. The victory is not inevitable.
IGTV SHOULD HAUNT THIS ANNOUNCEMENT
There is a reason Instagram has tried this before.
IGTV was launched in 2018 as the company’s ambitious long-form video bet. It was later discontinued as a standalone product, with Instagram moving away from that particular strategy.
The lesson is not that long-form video cannot work on Instagram. It is that adding a format does not automatically create a viewing habit.
Creators need an incentive to invest in longer productions. Viewers need a compelling reason to watch them. Advertisers need a credible way to reach those viewers. And the product needs discovery mechanisms that do more than serve existing followers.
Instagram now appears to be taking a more integrated approach, connecting long videos to Reels and developing experiences across mobile and television. That is more strategically coherent than simply launching another video destination.
Still, the graveyard of tech products is full of features that made sense in a presentation and failed to become habits.
THE MOST CONTROVERSIAL QUESTION: WHAT IF CREATORS ARE THE PRODUCT AGAIN?
Imagine Instagram successfully persuades thousands of creators to invest in 30-, 60-, and 90-minute productions.
They hire editors. They buy equipment. They spend days researching, filming, and polishing videos. Their audiences begin watching longer content on Instagram. Advertisers gain more inventory. Meta strengthens its position in the battle for television and digital video attention.
Now imagine that meaningful revenue sharing never arrives.
The platform would have secured a more expensive, more sophisticated content supply without accepting the financial obligations that made YouTube attractive to professional creators in the first place.
That is not a prediction. It is the risk creators should evaluate before reorganizing their businesses around a new feature.
The history of social platforms offers a familiar warning: distribution rules change, algorithms shift, and yesterday’s reach can disappear. Creators who mistake access to an audience for ownership of that audience are perpetually negotiating from weakness.
A platform can provide millions of views and still leave a creator financially fragile.
THE WINNER WILL BE WHOEVER PAYS FOR THE WORK
Instagram’s test is currently expanding from Brazil to a small group of users in the United States. That means the feature is early, and any declaration that YouTube is finished would be premature.
But the strategic direction is unmistakable. Instagram wants to stop being merely the place where viewers discover creators and become the place where they consume everything those creators make.
For creators, the decision is not whether to upload a 90-minute video. It is whether Instagram can offer enough incremental value to justify the production cost and the risk of platform dependence.
Test it. Measure watch time, retention, follower conversion, revenue per thousand views where available, and whether viewers return for subsequent episodes. Compare those results with equivalent content on YouTube. Do not confuse a spike in views with a durable business.
And do not abandon the channels, email lists, customer relationships, memberships, and revenue streams that you control simply because a platform has announced a new feature.
Instagram’s 90-minute video experiment could become a serious challenge to YouTube. But the real battle is not short-form versus long-form. It is subsidized attention versus sustainable creator income.
If Instagram builds transparent, competitive monetization around these videos, it could change where creators publish, where audiences watch, and where advertising money flows.
If it does not, this will be another masterclass in Silicon Valley’s favorite trick: announce a bigger opportunity, encourage creators to do more work, and leave the question of compensation for later.
Creators should welcome the opportunity. They should distrust the hype.
Because the most important number in this announcement is not 90 minutes.
It is zero: the number of dollars a creator is guaranteed to earn simply because Instagram allows a longer upload.
Until that changes, YouTube should pay attention. But creators should keep their eyes on the money.