Anthropic made its IPO prospectus public last week, but that filing only covered the first half of 2026. On Nov 2 it filed an amended Form S 1 with the SEC that adds results through Sept. 30, ahead of its road show. CNBC, Bloomberg and The Information all reported the new figures.

Zerohedge (@zerohedge) posted the headline figures from Anthropic's amended IPO filing on the morning of Nov 2. It reported third quarter revenue of $16.96 billion versus $1.38 billion a year earlier, total operating expenses of $17.5 billion versus $4.47 billion, $95.6 billion of cash and $36.3 billion of short term investments as of Sept. 30.

CNBC confirmed the filing. It reported that Anthropic generated about $17 billion in Q3 revenue, up about 1,127% from a year earlier, and $33.24 billion for the first nine months of the year, up roughly 1,200%. Growth slowed from quarter to quarter, though: Q3 revenue rose about 47% over Q2, after 144% growth in Q2 and 132% in Q1. Anthropic reported a Q3 net loss of $23.29 billion, and CNBC notes that most of the nine month loss came from changes in the fair value of convertible notes and derivative liabilities rather than operations. Bloomberg reported an operating loss of $567 million for the quarter, and other reporters noted that compute and infrastructure costs grew faster than revenue in Q3.

The filing lands as investors worry about the cost of the AI buildout. CNBC writes that Anthropic gets much of its capacity from Amazon, Google and SpaceX, and that rising interest rates and heavy hyperscaler spending have made markets nervous. The prospectus also includes more than 90 pages of risk factors. A separate story in this deck covers the $895.4 billion of compute commitments disclosed in the same filing.

Anthropic is expected to start its IPO road show soon, and investors will focus on whether revenue growth can keep pace with compute costs now that quarterly growth has slowed. Watch for a final share price range, the size of the offering and any October revenue update. The results will also shape how markets value OpenAI, which is reportedly aiming for a 2027 listing.

Lumentum makes lasers, optical transceivers and optical circuit switches used to link servers in AI data centers. Marvell designs custom AI chips and electrical and optical interconnects, and competes with Broadcom. Nvidia invested $2 billion in each company earlier in 2026, according to CNBC.

Unusual Whales (@unusual_whales) posted on Nov 2 that Marvell agreed to buy Lumentum in a $115 billion deal, calling it one of the biggest chip deals ever and a bet on optical components that connect AI data centers.

Marvell's own press release confirms the deal. Lumentum shareholders will receive 4.5 Marvell shares for each Lumentum share, an implied enterprise value of $115 billion for Lumentum, or about $1,132 per share. Marvell says that is a premium of about 23.9% to the closing prices on Oct. 28, the last trading day before market speculation began. After closing, Marvell shareholders will own about 67% of the combined company and Lumentum shareholders about 33%. Matt Murphy stays chairman and CEO, and three Lumentum directors, including CEO Michael Hurlston, join Marvell's board.

Marvell argues that combining its electrical and optical interconnect, switching and custom chip work with Lumentum's lasers and optical modules will let it design connectivity end to end. It says the combined company targets a $550 billion market and sees more than $1 billion in annual cost savings. CNBC frames the deal as Marvell stepping up its challenge to Broadcom, while Bloomberg, the Financial Times and The Wall Street Journal also reported it. Axios' Dan Primack suggested it could be the largest tech merger ever, excluding related party deals.

The companies expect the deal to close in the first half of 2028, subject to shareholder and regulatory approvals, and analysts have flagged China's review as a possible hurdle.

The long timeline means regulators in the US, Europe and especially China will have plenty of time to weigh in, and any condition could reshape the deal. Watch how Broadcom, Coherent and other optics suppliers respond, and whether Nvidia's stakes in both companies draw scrutiny. Shareholder votes and early integration plans will show whether investors buy the end to end design pitch.

Anthropic's amended IPO filing on Nov 2 says it has purchase commitments and lease agreements with an aggregate committed spend of about $895.4 billion, substantially all of it for compute and infrastructure. CNBC confirmed the figure. The comparisons with Google and Meta come from zerohedge's own analysis of off balance sheet liabilities.

Zerohedge (@zerohedge) wrote on Nov 2 that Anthropic has $895.4 billion of committed spend according to its latest S 1, up from about $105 billion at the end of 2025. It added that the only hyperscaler with more is Google at $1,059 billion, with Meta at $866 billion.

The headline number is confirmed by CNBC, which quotes the filing as saying Anthropic has entered into purchase commitments and lease agreements with an aggregate committed spend of about $895.4 billion, substantially all of which is related to compute and infrastructure. CNBC also reports that compute and infrastructure expenses rose $14.4 billion in the first nine months of the year, up 329%.

Zerohedge's chart breaks the total down. It shows $313.8 billion of purchase commitments, $451.2 billion of leases that have not yet started and $106.8 billion of commitments signed after Sept. 30, among other items. A hatched bar marks another $150 billion to $175 billion, which the chart attributes to additional Google TPU capacity and AMD Helios systems with no fixed dollar commitment, so it is not in the total. Zerohedge followed with charts comparing Anthropic's off balance sheet liabilities with hyperscalers and listing commitments by counterparty. The hyperscaler comparisons are zerohedge's own calculations and were not checked line by line here.

For context, the same filing shows Q3 revenue of about $17 billion, so the commitments are roughly 13 times a single quarter's sales.

Investors will study how much of the $895.4 billion is cancellable, when the leases start and which suppliers carry the most exposure. Watch for the road show to address how Anthropic plans to pay for these commitments if growth keeps slowing. Suppliers like Google, Amazon and AMD may also be asked how much of their own backlog depends on Anthropic.

Meituan, the Chinese food delivery and local services company, builds the LongCat family of models. LongCat 2.5 follows LongCat 2.0 and is described by Meituan as natively multimodal and built for long horizon agentic tasks. Meituan says it trained and serves the model on domestic AI accelerators.

TestingCatalog (@testingcatalog) posted on Nov 2 that Meituan released LongCat 2.5, a multimodal model with 1.6 trillion total parameters and 49 billion active, compared with GPT 5.5, GPT 5.6 Sol, Claude Opus 4.8 and Claude Opus 5, and now available on the APIs.

Meituan's own launch post and blog give the active count as approximately 48 billion per token, slightly lower than TestingCatalog's figure. Meituan describes LongCat 2.5 as a natively multimodal agentic model for long horizon tasks, with stronger visual understanding and agentic coding than LongCat 2.0. It turns images into discrete tokens so the same backbone handles text and vision, and it was trained with large scale reinforcement learning on long coding tasks. Meituan says it optimized long context training and 4 bit inference on domestic AI accelerators.

All benchmark numbers come from Meituan and are its own claims. Its table reports, for example, 88.4 on Terminal Bench v2.1, 91.9 on BrowseComp and 63.1 on Humanity's Last Exam with tools. The same table shows Claude Opus 5 ahead on several tests, including Terminal Bench v4.0, where Meituan lists 53.9 for Opus 5 against 38.5 for LongCat 2.5. Meituan notes that scores for other models marked in its table come from public reports, while others were run by Meituan and are indicative only.

Independent evaluations have not yet been published, so how LongCat 2.5 performs outside Meituan's own tests is still unknown.

Watch for independent benchmark results and leaderboard entries over the coming weeks, which will show whether Meituan's claims hold. Developers will test pricing and availability on its API, and whether open weights follow as with earlier LongCat models. The domestic chip training claims will also draw attention as US export controls continue. If the results hold up, LongCat 2.5 would add pressure on other Chinese labs and on Western providers to cut prices for agentic coding.

GLM 5.3 is an open weight model from Zhipu's Z.ai. FrontierScience is a benchmark of difficult science questions, with a Research track meant to test open ended research problems, and GPQA Diamond is a widely used set of graduate level science questions. Chenru Duan (@chenru_duan) is a researcher leading the Hydrogen team.

Chenru Duan (@chenru_duan) announced on Nov 2 Hydrogen 1.0, which the post calls the team's first open weight model for scientific research.

According to the post, the team post trained GLM 5.3 from Z.ai (@Zai_org) on its internal scientific data, lab records and discovery trajectories. It says Hydrogen 1.0 scores 38.3% on FrontierScience Research against GPT 6 Astra's 33.3%, and 93.0% on GPQA Diamond. These figures are the team's own claims and had not been reproduced by an independent evaluator at the time of posting. The post does not say whether GPT 6 Astra was run with the same settings or tools.

The post makes an argument for openness. It says science should be open, collaborative and democratized rather than proprietary, and that only open weights let researchers inspect, adapt and build on the speed of AI progress. It notes that several recent models are named after chemical elements, but says Hydrogen 1.0 is the first aiming toward an open scientific superintelligence for discovery.

The result, if it holds, is notable because FrontierScience Research is meant to measure harder, open ended tasks where closed frontier models have led. A post trained open model beating one of them would show how much domain data and research trajectories can add on top of a strong open base. It also extends the trend of Chinese open models such as GLM serving as the base for specialised Western and academic projects. The thread promises more details, including where the weights can be downloaded.

Watch for the weights, license and technical report, and for independent scores on FrontierScience and GPQA Diamond. Labs testing it on real research tasks will show whether the benchmark lead holds up. If it does, expect more science models built on open bases, and pressure on closed labs to show stronger research results of their own. Z.ai's own response, and whether it promotes Hydrogen as a GLM success story, is also worth watching.

In September OpenAI said a large swarm of AI agents produced a proof about blowup in the Navier Stokes equations, one of the Millennium Prize problems, with a formal Lean check. Mathematicians have debated how much it settles, since the result relies on a carefully chosen external force. On Oct 31 Altman said OpenAI let its AI math work run for a week.

Prinz (@deredleritt3r) posted on Nov 2 a clip of OpenAI CEO Sam Altman talking about the internal model behind OpenAI's Navier Stokes result. The post says Altman confirmed the model will be released soon.

The quote in the post reads: It was 10,000 agents of the model that is going to be released soon, the next GPT, working for about four days, nonstop, collaborating with each other perfectly. This is secondhand: the quote comes from prinz's transcription of a linked clip, and the post does not say where or when Altman spoke. The quote does not give a name, a date or what soon means.

The background matters. In September OpenAI announced that agents running a model beyond GPT 6 Astra produced a proof showing singular behaviour in the Navier Stokes equations under a smooth but extreme external force, and released a formal Lean verification. Some mathematicians praised the work, including PDE expert Scott Armstrong, as covered in the Oct 25 deck, while critics argued the setup sidesteps the physical version of the problem. OpenAI did not claim the Millennium Prize.

What is new is the link to a product. Until now the math work was described as coming from an internal model. If Altman's words are accurate, the same model will ship as the next GPT, and the result shows OpenAI running it at very large scale with thousands of cooperating agents over days. A public release would let outsiders test how much of that ability holds without OpenAI's internal setup.

Watch for OpenAI to name and date the next GPT, and for whether it ships with the multi agent tools used in the math work. Independent tests of long research tasks will show how much of the Navier Stokes ability carries over. The original clip and its venue should also become clearer as more outlets pick up the quote. Mathematicians are still checking the proof, so their verdict will shape how much weight the claim carries.

Oracle outlined the raise at an investor day last week, where it also forecast at least $225 billion of revenue for fiscal 2030 and said it would limit new capital market funding for the build out to $20 billion. On Nov 2 it filed to sell $10 billion of common stock and $10 billion of senior notes. CNBC, Bloomberg and Reuters reported the filings.

The Kobeissi Letter (@KobeissiLetter) reported on Nov 2 that Oracle stock fell over 5% after it announced plans to raise $20 billion through equity and debt to fund its AI buildout, adding that the stock is down 54% from this year's high.

CNBC confirms the filings: Oracle will sell $10 billion of common stock and $10 billion of senior notes, and shares tumbled more than 5% on Monday. Reuters reports that the bond offering has six tranches maturing between 2029 and 2046. Oracle said it will use the money for general corporate purposes, which may include capital spending and debt repayment.

The plan was not a surprise. Oracle laid it out at last week's investor day, alongside a fiscal 2030 revenue target of at least $225 billion and a pledge to cap additional capital market funding for this investment cycle at $20 billion. The stock briefly rose after that event. CNBC notes that Oracle had $125.34 billion of debt as of Aug. 31, that S&P cut it to BBB minus in July, one notch above junk, and that its five year credit default swap spread is up about 270% in a year. Oracle says vendor financing, customer supplied hardware and prepayments will help it reach positive free cash flow in the second half of 2028.

The 54% figure is Kobeissi's own measure from the yearly high. CNBC puts the decline at 33% so far in 2026 and 50% over the past 12 months.

Watch the pricing of the bond tranches and the share sale, which will show how much investors charge Oracle for AI risk. Any sign that Oracle needs more money before 2030 would break its own pledge and could hit the stock again. Its next earnings report and updates on big customer contracts will test whether the revenue target looks realistic.

Dots are OpenAI's agents that each get their own cloud computer and browser, launched in October and powered by GPT 6 Astra for eligible Pro and Business Premium users. GPT 6.1 Terra is the faster, lower cost model in OpenAI's GPT 6.1 family, released in late October. Tibo (@thsottiaux) works at OpenAI and often posts product updates.

Tibo (@thsottiaux) posted on Nov 2 a note addressed to fellow resetters, a nickname used among dots users, saying OpenAI is working on expanding the family of frontier models that power dots. He names GPT 6 Astra and GPT 6.1 Terra, and says some others will follow in the future.

He gives one reason for adding Terra: it brings speed and a nice personality upgrade, and will help OpenAI bring dots to more people. The post does not give a date, pricing or which plans will get the new models.

According to OpenAI's October release notes covered earlier in this archive, dots are powered by GPT 6 Astra and each has its own cloud computer and browser. They have been rolling out to eligible Pro and Business Premium users aged 18 and older, with Enterprise access in beta. Usage was not counted toward plan allowances for the first month after launch.

GPT 6.1 Terra was released in late October as a faster, cheaper member of the GPT 6.1 family. Public posts and benchmark summaries describe it as aimed at computer use and agentic tasks at about half the price of GPT 6.1 Sol, though those comparisons come from third party tests rather than this post. Pairing a cheaper model with dots fits Tibo's stated goal of wider access, since each dot runs long tasks on its own machine and model cost is a large part of what it costs to serve.

This is an announcement of work in progress, not a launch.

Watch for OpenAI release notes that add a model picker or Terra option to dots, and for any change in which plans can use them. Pricing will matter once the free first month ends. If Terra handles everyday dot tasks well, OpenAI could open dots to Plus users or more markets. Tibo's mention of other models suggests more options may follow.

Abu Dhabi has become one of OpenAI's closest partners. Its fund MGX co led OpenAI's $122 billion funding round, according to The National, and G42 is a key AI infrastructure partner in the UAE. Sam Altman met Crown Prince Sheikh Khaled bin Mohamed bin Zayed on Nov 2, who witnessed the signings.

Reuters (@Reuters) reported on Nov 2 that Abu Dhabi's ADNOC and G42 signed a deal with OpenAI as Sam Altman traveled to the UAE.

The Abu Dhabi Media Office says Crown Prince Sheikh Khaled bin Mohamed bin Zayed met Altman and witnessed four agreements between Abu Dhabi entities and OpenAI, covering government, energy and investment. In the first, ADNOC, G42 through its subsidiary Inception42, and OpenAI agreed to deploy OpenAI Frontier across ADNOC's operations, which ADNOC says makes it the first energy company to use the platform. It will run on GPT models with inference infrastructure inside the UAE. ADNOC says the first uses are planning whole oil field developments at once and forecasting supply, demand, prices and margins across its value chain, with humans keeping oversight of major decisions.

The other three agreements are a Machine Readable Government initiative with the Abu Dhabi government, which turns a set of policies and rules into a form AI systems can use for routine compliance checks such as permits; a deal between Mubadala, Inception42 and OpenAI to apply AI to investment analysis and risk assessment; and a five year agreement between Inception42 and OpenAI to support AI adoption across government, enterprise and education in the UAE.

Altman said Abu Dhabi and the UAE are setting a bold global standard for AI integration across the public and private sectors, according to The National. No financial terms were disclosed.

Watch for the first results from the ADNOC deployment and from the government pilot, which The National says started with food import rules. The deals could lead to more UAE compute built for OpenAI and closer scrutiny in Washington of AI partnerships in the Gulf. Other energy giants may follow with their own frontier AI contracts, and OpenAI may announce more Gulf partnerships during Altman's trip.

Hume AI builds voice AI focused on emotion and expression, including its empathic voice interface and text to speech models. Understanding audio usually takes several specialised models, such as speaker counting, emotion detection and event detection. The Audio Decision API is offered through a request access form rather than general release.

Hume AI (@hume_ai) introduced the Hume Audio Decision API on Nov 2 with a launch video showing its team at work.

The post says users can ask anything about audio, giving examples such as how many speakers there are, whether anyone is laughing and whether the caller is frustrated. Hume says what used to take more than 10 models is now one call. It says the system was trained on more than 10 million human judgments, so it answers the way human listeners would, and that it works in realtime or across hours of recordings. Access is by request.

These are Hume's own descriptions. The post does not give accuracy figures, supported languages, latency or pricing, and no independent tests were available at launch.

The pitch is a question answering layer for sound. Instead of wiring up a speaker diarization model, an emotion classifier and an event detector, a developer would describe what they want to know in plain language and get an answer. That fits a broader shift toward general multimodal models that handle many tasks from one prompt, applied here to an area where Hume has long focused on vocal expression.

Likely uses include customer support monitoring, meeting analysis, media review and safety tools that flag distress. Those same uses raise privacy questions, since judging a caller's emotional state from their voice is sensitive, and the post does not say how consent or data retention are handled.

Watch for pricing, documentation and accuracy numbers once access widens, and for early customers in call centers or meeting tools. Comparisons with general audio models from OpenAI and Google will show whether a specialised API keeps an edge. Privacy rules on emotion recognition, especially in Europe, could shape where it can be used. Hume's own voice products may also start using the API to react to how users sound.

DeepSeek Harness is DeepSeek's agent app. On Nov 1 it open sourced a proactive coding agent prototype, the top story of that day's deck. Creator Mode is the next feature it has announced, aimed at letting users define their own tools inside the app.

DeepSeek Harness (@DeepSeekHarness) posted on Nov 2 a launch video for Creator Mode under the line Your Harness. You Define It.

The post says Creator Mode turns ideas into plugins that you can use, and gives three examples: a design workspace, a sandbox game and a 3D LLM visualizer. It closes by asking users what they will create and links to more information. The video runs about 100 seconds and opens on the DeepSeek Harness logo.

The post does not explain how plugins are built, whether they are generated by the agent from a plain language description, which models power the feature, or whether it costs extra. It also does not say whether plugins can be shared with other users. Those details should come from DeepSeek's own documentation rather than from the launch post.

The launch follows a busy weekend for the account. On Nov 1 DeepSeek Harness open sourced a prototype of a proactive coding agent, which became the top story of that day's deck. Creator Mode points in a different direction: instead of only helping developers write code, the app is pitching itself as a place where users shape the tool itself by describing what they want.

The idea is not new across the industry, since several AI apps let users build small apps or artifacts from prompts. The difference here is the framing as plugins inside an agent harness, which suggests the creations are meant to plug back into the agent's workflow rather than stand alone. The 600K views show strong interest in DeepSeek's agent products.

Watch for documentation showing how Creator Mode plugins are built, stored and shared, and whether there is a public gallery. Usage limits and pricing will show who it is for. If users start sharing useful plugins, DeepSeek Harness could become a platform rather than a single agent. Rival agent apps may respond with similar features. Security reviews will also matter, since plugins that run inside an agent can reach files and tools on a user's behalf.

As recently as July, Goldman's chief information officer Marco Argenti praised vibe coding by traders and bankers who built tools with AI. Goldman also runs its own internal AI assistant that combines models from OpenAI, Anthropic and Google. Earlier this year it restricted Claude for staff in Hong Kong.

Bloomberg (@business) reported on Nov 2 that Goldman Sachs has irked some of its staff after banning most of them from using advanced AI coding tools, following a review of productivity and risk concerns.

According to Bloomberg's Todd Gillespie, Goldman had for months let bankers use Anthropic's Claude Code and Microsoft's GitHub tools for work. Internal guidance updated in October now limits those tools to engineers. A person familiar with the matter told Bloomberg that some traders are frustrated, because they had used the tools to monitor client trading flows, model market changes and positions, and create written and graphic content. The ban does not apply to strats, the staff who work between trading, quantitative research and software engineering.

Goldman spokesperson Nick Carcaterra told Bloomberg that the bank makes certain AI tools available to find the best use cases, and after a period of experimentation that weighs factors including productivity and risk management, puts them in the hands of those best suited to use them long term.

The move contrasts with the bank's earlier enthusiasm. In July, Argenti described non technical staff building solutions with AI and handing them to the firm's 12,000 engineers to improve. Bloomberg places the change within a wider effort on Wall Street to work out how thousands of workers should use AI while managing cybersecurity, cost and productivity. The report relies on internal guidance and unnamed sources, and Goldman's statement did not dispute it.

Watch whether other banks follow with similar limits or keep widening access to coding agents. Anthropic and Microsoft may face more questions from large customers about controls, audit trails and data access. If Goldman later reopens the tools with stricter guardrails, that could become a template for regulated industries. The frustration among traders will also show how hard it is to take back tools once staff rely on them every day.

Fable is a Claude model line, and unconfirmed leaks since mid October have said a Fable 5.5 checkpoint is being tested. Anthropic has not announced Fable 5.5. Chetaslua (@chetaslua) is an AI tester who posts early model demos, and Chubby (@kimmonismus) reshared the video.

Chetaslua (@chetaslua) posted on Nov 2 a short animated film he says was made entirely in code by Fable 5.5, writing that it is over for video and image models. He says the model one shotted the video.

In a follow up post, he gave more detail. He says Fable 5.5 made a 66 second animated short with zero image models, and that every frame and every sound is Python the model wrote itself, including a fox, paper birds, clouds, lettering, a score and sound effects. He lists the output as 1920 by 804 pixels at 24 frames per second in a single HTML file, with no generated images, stock assets, samples or font files, and about 2 hours 40 minutes of autonomous work. He adds that it is not Ghibli level, and that the model told him so.

The key question is which model made it. Chetaslua offers two tests: ask the model where Andrej Karpathy went in 2026 with web search and memory off, which he says only the new Fable knows, or run the same prompt shown in his screenshot. He says regular Opus 5.5 produces lower quality and Fable 5.1 is worse. Chubby (@kimmonismus) reshared the video and called it most likely the latest Fable 5.5 checkpoint.

None of this is confirmed. Anthropic has not announced Fable 5.5 or said which models are in testing, the knowledge test is indirect, and the code itself has not been published for others to check. The video is real, but the model behind it is a claim by the poster.

Watch for Anthropic to announce Fable 5.5 or publish a model card, and for other testers to reproduce the film from the same prompt. Publishing the HTML file would let others confirm it is pure code. If confirmed, expect more demos of coding models producing animation, games and music, and new debate over how they compare with dedicated video models.

Moonshot launched Kimi K3.1 Pro in late October and said its weights would be released on Nov 4. In recent days posters have alleged that some praise for the model came from undisclosed paid promotion. WhatsApp's account posted kimi, you made history on Nov 1, a post that passed 3.9 million views.

Chubby (@kimmonismus) posted on Nov 2, quoting a WhatsApp post, that even WhatsApp knows about the controversy over undisclosed promotion of Kimi's K3.1 Pro model, and called it embarrassing for Kimi.

The post he quotes is from WhatsApp's own account on the evening of Nov 1 and reads only kimi, you made history, with a trophy emoji. It passed 3.9 million views. WhatsApp did not explain the post, and it is Chubby's reading that it refers to the promotion controversy. The post could also be read as a general joke, so its meaning is not confirmed.

The underlying allegation is that some of the online praise for Kimi K3.1 Pro came from paid promotion that was not disclosed as advertising. Posts summarising the dispute describe these as allegations. Moonshot had not publicly responded in posts found for this story, and no evidence of specific paid posts was reviewed here, so the claim should be treated as unproven.

The model itself has done well on some public measures. Earlier stories in this archive covered its launch with weights due Nov 4, its placement at number 11 in Arena's Agent Arena, and Moonshot resetting quotas after heavy demand. One posted summary puts it at 55 on Artificial Analysis's intelligence index, though that comes from a low reach account.

The story is mostly about trust. If paid promotion shaped early impressions, developers may wait for independent tests once the weights ship.

Watch for Moonshot to respond to the allegations and for whether the Nov 4 open weight release goes ahead on schedule. Independent benchmarks run on the released weights will matter more than social posts. X's paid partnership disclosure rules could come up if specific undisclosed posts are identified. Other labs may also face closer scrutiny of influencer posts around their launches, and Arena and Artificial Analysis rankings will be watched for signs of manipulation.