The AI race is not slowing down.
Claude, GPT, Gemini, Grok and Qwen all shipped new models in the same week. Anthropic's draft S-1 still carries no ticker, ten startups raised, and two new ETFs opened doors a reader can actually walk through.
00 THE SIGNAL
WHAT IS IN THIS ISSUE
The AI race is not slowing down. The answer is in: Claude, GPT, Gemini, Grok and Qwen all shipped new models inside the 19–26 September window.
Claude Opus 5.5 shipped on 22 September, the first model in the 5.5 line. GPT-6 Sol and GPT-6 Luna shipped the same day. Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS shipped on 23 September. Grok 4.7 shipped on 21 September. Qwen3.8-Omni-Flash shipped on 20 September and Qwen-Image-2.1 shipped on 21 September. Section 01 takes them one at a time, with the page we opened.
Nothing left hanging. This week is measured by what shipped, not by what is promised for next week.
WINDOW. This issue covers 19–26 September. The tape has nine cells. Section 02 has two machines that were announced but did not go on sale inside the window.
IN THIS ISSUE
- Models this week — Claude Opus 5.5; GPT-6 Sol; GPT-6 Luna; Gemini 3.8 TTS; Grok 4.7; Qwen-Image-2.1; Qwen3.8-Omni-Flash
- Not AI, for once — the Surface Laptop 13-inch and Surface Pro 12-inch, announced by Microsoft on 23 September, on sale from 13 October
- Where the money went — BDACS; BDx; IBM. Nine cells on the tape
- The star — Anthropic and an IPO still waiting. No listing inside the window
- The raises — Heidi $340M; Firecrawl $75M; Chamelio $26M
- Research radar — Anthropic on economic scenarios; HarnessTax; Chainalysis
- Access — Coinbase opens Oura IPO allocations to US customers; two new ETFs list. Anthropic's door is not open yet
- CAL ecosystem — the two company cards, and four questions for Alex
- What's next — OpenAI DevDay 29 September · GITEX Vietnam 1–2 October · World AI Week 5–9 October · FOMC 27–28 October
01 MODELS THIS WEEK
WHAT SHIPPED
CLAUDE OPUS 5.5 · 22 SEPTEMBER
Claude Opus 5.5 is the first model in the Claude 5.5 line. Anthropic says it handles most work on a par with Claude Fable 5.1, and that at default settings it runs about 40% cheaper than Opus 5 on typical loads in the company's own testing. List price: $4 per million input tokens, $20 per million output tokens. Sonnet 5.5 and Haiku 5.5 are described by the company as coming in the weeks ahead, not this week. · anthropic.com/claude-opus-5-5
GPT-6 SOL · 22 SEPTEMBER
TechCrunch, 22 September, 11:00 Pacific: OpenAI shipped GPT-6 Sol for complex work such as programming. The company says API pricing for the 6 line is half that of the Sol and Luna 5.6 line, owing to caching and inference. Sol is in ChatGPT Work and Codex for most paid accounts, and in the API. · techcrunch.com
GPT-6 LUNA · 22 SEPTEMBER
Same article: OpenAI shipped GPT-6 Luna for high-volume work with a clear target: summarising, extracting information, quick questions. Luna is in the API, in ChatGPT Work and Codex for most paid accounts, and in the desktop app for Free and Go users. TechCrunch quotes the company: on an internal factuality test, GPT-6 Sol makes about half as many errors as the previous version. · techcrunch.com
GEMINI 3.8 FLASH TTS · 23 SEPTEMBER
Google, 23 September: Gemini 3.8 Flash TTS for directing voice and building characters. It creates new voices from a written description, and controls pacing, regional accent and line-by-line delivery. Rolling out in the Gemini API, Google AI Studio and Google Vids. The API for Gemini Enterprise is listed as coming. · blog.google
GEMINI 3.8 FLASH-LITE TTS · 23 SEPTEMBER
Same page, same day: Gemini 3.8 Flash-Lite TTS for high volume, dubbing, audio content and voice agents. Also rolling out in the Gemini API, Google AI Studio and Google Vids. The text version of Gemini 3.8 Flash shipped on 2 September, outside the window, and is not counted here. · blog.google
GROK 4.7 · 21 SEPTEMBER
The x.ai page is dated 21 September 2026. Grok 4.7 uses a larger base model than Grok 4.6, at the same price and serving speed as Grok 4.6: from $2 per million input tokens and $6 per million output tokens. Available in Cursor, Grok Build and the Grok API. A fast version runs at double the output speed for double the price. · x.ai/news/grok-4-7
QWEN3.8-OMNI-FLASH · 20 SEPTEMBER
Alibaba Cloud, 20 September: Qwen3.8-Omni-Flash takes text, images, audio and video, with a 1 million token context window, on the Qianwen AI Platform. The page states that the API price per hour of input audio is down more than 98% against their own earlier estimate. · alibabacloud.com
QWEN-IMAGE-2.1 · 21 SEPTEMBER
Alibaba Cloud, 21 September: Qwen-Image-2.1 is open weights, a model that both generates images from text and edits images, with an alpha channel. The generator: 7 billion parameters, 32 Single-Stream DiT layers. Up to 10 reference images. · alibabacloud.com
02 NOT AI, FOR ONCE
NEW HARDWARE
MICROSOFT · SURFACE LAPTOP 13 INCH
Blog Microsoft Devices, 23 September 2026: announcing the Surface Laptop 13-inch on Snapdragon X2 Plus, a thin Windows machine with a full keyboard. Starting at $1,199 MSRP. On sale from 13 October in selected markets — not on sale inside the 19–26 September window. · blogs.windows.com
MICROSOFT · SURFACE PRO 12 INCH
Same article, same day: the Surface Pro 12-inch, a fanless Windows machine that goes from tablet to laptop when the keyboard is attached. Starting at $1,149.99 MSRP. The on-sale date is 13 October, so not this week. Keyboard and Slim Pen sold separately. · blogs.windows.com
The event inside this week is the 23 September announcement. Both machines go on sale on 13 October, so neither was on sale inside the 19–26 September window.
03 WHERE THE MONEY WENT
DEALS & INFRASTRUCTURE
BDACS · FIREBLOCKS FOR KRW1
BDACS, 21 September, picked Fireblocks as technical partner for institutional custody and stablecoin transaction infrastructure. The article states that the Prime Custody Solution passed 80 billion won ($58 million) in assets under custody as at the first half of this year. KRW1 is connecting Ethereum, Avalanche, Circle ARC, Polygon and Aptos. Same article: Fireblocks has supported more than $16 trillion in digital-asset transfers across more than 200 chains. · bloomingbit.io
BDX · A POWER CAMPUS FOR AI
BDx Data Centers, 22 September, broke ground on AI Campus 2 (CGK4) at Jatiluhur, West Java: a 640 MW campus. The PLN grid supplies 845 MVA, part of more than 1.2 GVA of power BDx already holds in Indonesia. Building 1: 120 MW of IT capacity, with the first contracted phase expected in service in early 2027, and direct cooling up to 500 kW per rack. The page also states a $320 million credit facility already in place during 2026, with no signing date inside this week. · siamnewsnetwork.net
IBM · DIGITAL ASSET HAVEN
IBM, posted 24 September: two steps for Digital Asset Haven. One, connecting financial institutions to Swift's blockchain-style shared ledger, with a beta of the ISO 20022 message translator. Two, an on-premises beta on IBM Z and LinuxONE, inside the customer's own Red Hat OpenShift, with no dependence on public cloud. Swift is designing the ledger with more than 40 financial institutions. The IBM page states no monetary figure. · ibm.com
THE TECH TAPE · NINE PRICES, SEVEN DAYS
Nine prices on the tape. Eight are closes for the 25 September 2026 session, read against the 18 September session on the same series. Samsung closes on 23/09/2026 against 18/09/2026. The US equities and SpaceX Class A (ticker SPCX) come from the Yahoo Finance chart; the session bar is stamped 25/09/2026 New York time. Ethereum is not a listed equity; the day point comes from CoinGecko history, labelled 25/09/2026 against 18/09/2026. Percentages are computed from those two prices.
- Nvidia $225.07 · ▲ +1.26%
- TSMC $450.61 · ▲ +3.67%
- Apple $341.07 · ▲ +1.47%
- Microsoft $516.17 · ▲ +4.53%
- Alphabet $343.92 · ▼ −1.61%
- Ethereum $2,687 · ▲ +9.85%
- Samsung ₩286,500 · ▲ +10.19%
- SpaceX $148.68 · ▼ −2.64%
- Meta $751.66 · ▲ +12.90%
23/09 was the last session of the week, so that is Samsung's figure for the week.
BEST OF THE WEEK
- 1 · Intel $123.00 · ▲ +13.26%
- 2 · Meta $751.66 · ▲ +12.90%
- 3 · AMD $630.63 · ▲ +12.65%
04 THE STAR
ONE RESULT, CLOSE UP
ANTHROPIC · A DRAFT S-1, NO TICKER YET
Anthropic is an AI research and safety company. It says it builds systems that are reliable, interpretable and steerable. The product that carries the research outward is Claude. Anthropic is a public benefit corporation. The purpose stated on the company page: to responsibly develop and maintain advanced AI for the long-term benefit of humanity. Board: Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell, Vas Narasimhan. Long-Term Benefit Trust: Neil Buddy Shah, Richard Fontaine and Ben Bernanke. · anthropic.com/company
This week the company shipped Claude Opus 5.5, covered in section 01. This piece is a portrait of the company; it does not repeat the model's scorecard.
The IPO is still a thing waiting to happen, not a thing that has happened. On 1 June 2026, Anthropic, PBC confidentially submitted a draft Form S-1 to the SEC for a proposed offering of common stock. The company says the number of shares and the price range have not been set, and that the offering depends on market conditions and other factors. That announcement is not an offer to sell. · anthropic.com
Reuters, 4 September 2026, updated 8 September, citing people familiar with the matter: marketing of the IPO is expected no earlier than mid-October, and a public prospectus was, at that point, not expected before the end of September. Reuters notes that plans, including timing, can change. Anthropic declined to comment. There is no public S-1, no ticker, no exchange and no listing date inside 19–26 September 2026. The buying door is not open, so there is no line for it in section 07. · reuters.com
No valuation, share count or ticker is printed here: the company's own filing does not carry those numbers.
OPEN SOURCE WORTH AN HOUR
- mattpocock/skills — a working engineer's agent skills, published straight from the author's own .agents directory. Shell, MIT. 270,409 stars; last push 24 September.
- deepseek-ai/deepseek-harness — DeepSeek's own agent harness, built on one rule: everything is a plugin. TypeScript, MIT. 237,220 stars; last push 24 September.
- microsoft/markitdown — turns PDFs, Office files and the rest into Markdown, which is the form an agent can actually read. Python, MIT. 187,219 stars; last push 21 September.
05 THE RAISES
STARTUP FUNDING
HEIDI · 22 SEPTEMBER
Heidi, 22 September, closed a Series C of $100 million led by Blackbird, at a valuation of $900 million. Alongside it, a growth tranche of $240 million led by the General Catalyst Customer Value Fund, for total new capital of $340 million. The money goes to pushing Heidi deeper into health systems, adding agent capability around the consultation, and to the clinical-evidence and regulatory work. ARR went from $1 million to $50 million, as at April 2026. The platform has supported more than 175 million consultations, about 2.8 million a week, across 190 countries and 110 languages. Heidi Evidence has answered more than 10 million queries since March 2026. The company has shipped its first wearable, Remote, capturing audio from the operating theatre to the community clinic. Upcoming product capability will not be available in the UK or the EU. · heidihealth.com
FIRECRAWL · 22 SEPTEMBER
Firecrawl, 22 September, Series B $75 million led by Smash Capital, with Altos Ventures, Nexus Venture Partners, Y Combinator, Freestyle and Offline Ventures. The round opens Alexandria: one door through which an agent can see a source, know what that source holds, and then pull from the live web, Firecrawl's own index, data providers and private connectors. Part of the money goes to paying people and organisations when agents use their knowledge. The company already pays Wikimedia Enterprise; millions of Wikipedia data requests pass through Firecrawl each month. The Research Index holds tens of millions of paper abstracts. The Developer Index holds tens of millions of documentation sources, READMEs, issues and pull requests. The Government Index covers laws and regulations. More than 1.5 million users. · firecrawl.dev
CHAMELIO · 22 SEPTEMBER
Chamelio, 22 September, Series A $26 million led by Entrée Capital, with Work-Bench, Emerge Ventures and Bright Pixel Capital. The money goes to developing its own legal action model, growing the legal-engineering and product teams, and further investment in the technology that onboards customers. The software drafts from the company's own contracts, reviews and negotiates terms, routes requests and manages obligations. ARR is up fourfold since the seed round five months ago. Customers number in the hundreds, including Wiz, monday.com, Socure, AppsFlyer and Wonderful.ai. · prnewswire.com
NUMERAL · 23 SEPTEMBER
Numeral, 23 September, Series C $100 million led by Insight Partners, with Salesforce Ventures, Geodesic, Benchmark, Mayfield, FCVC, Y Combinator and Uncork. The money goes to pushing the product, expanding into software, manufacturing, distribution and wholesale, and growing the engineering, sales, marketing and product teams. The platform handles sales tax along with VAT and GST in more than 90 countries, connecting more than 40 billing systems and ERPs. The page states $157 million raised to date. · numeral.com
EMA · 23 SEPTEMBER
Ema, 23 September, Series B $77 million led by Creaegis; Accel, S32 and Prosus increased their existing positions. Total capital $140 million. The valuation is more than four times the previous round. The money goes to building out the sales organisation and expanding into APAC and EMEA. The platform runs AI employees for enterprise operations. At Wipro it assists more than 240,000 staff across 65 countries and more than 100 processes, handling about 2.9 million questions a year. · ema.ai
TEKEVER · 23 SEPTEMBER
TEKEVER, 23 September, closed a first tranche of a Series D of $580 million, at a valuation of $6.4 billion. UC Investments and Baillie Gifford led. Merlyn Advisors is new; Crescent Cove, Ventura Capital and Iberis Capital continued. The company says this tranche covers most of the round, with later tranches still open. The money goes to international presence, industrial and technological capacity, acquisitions, and AI-driven autonomous systems. Same page: the UK Ministry of Defence selected TEKEVER to deliver CORVUS, built on the AR5 system, a programme worth up to £400 million over ten years. · tekever.com
ENVEDA · 23 SEPTEMBER
Enveda, 23 September, closed a Series E of $311 million led by Catalio Capital Management. Total capital since founding is more than $845 million. The money goes to taking ENV-294 and ENV-308 into later-stage trials, pushing ENV-6946 through mid-stage, moving further inflammation and metabolic drugs into clinical trials, and expanding PRISM together with the automated laboratory that feeds that platform. Headquartered in Boulder. · enveda.com
ISLAND · 24 SEPTEMBER
Island, 24 September, Series F $400 million, at a valuation of $6.4 billion, more than double the 2024 mark. Evolution Equity Partners led. The money goes to a growth phase as enterprises put AI agents to work: the layer that sees, controls and audits. The product moves from the Enterprise Browser to a control plane for both people and agents, across browser, endpoint, network and data. The company has 1,000 staff. ARR has doubled every fiscal year since launch in 2022. · island.io
BIGHAT BIOSCIENCES · 24 SEPTEMBER
BigHat Biosciences, 24 September, completed a Series C of $75 million. Total raised is $223 million. DFJ Growth and Premji Invest co-led. The money goes to the platform that generates its own biological data for protein design, and to pushing a pipeline including BHB810 and BHB299 toward clinical readouts. Headquartered in San Mateo. · bighatbio.com
KONTEXT · 24 SEPTEMBER
Kontext, 24 September, raised $4 million led by 42CAP, with a16z CSX and HTGF. The money goes to growing the engineering team and continuing the runtime-enforcement platform that lets customers deploy agents with controls in place. The software sits between agents and the tools they call, checking activity against policy before an action runs. Headquartered in Munich. · kontext.security
06 RESEARCH RADAR
AI IN USE
HARNESSTAX · SAME MODEL, DIFFERENT BILL
21 pairs (7 models × Claude Code, Codex CLI, Pi) across 30 SWE-bench Lite tasks and 30 Terminal-Bench 2.0 tasks, three runs each. Published 16 September; AlphaSignal carried it on 20 September. On SWE-bench Lite, Claude Fable 5 resolved 97.8 percent of attempts with Claude Code and 96.7 with Pi, with overlapping confidence intervals, at $1.33 against $0.67 per attempt — the figures on the source page. Claude Code's first-call context, averaged across seven models, is more than ten times Pi's. Across six Anthropic and OpenAI models and both benchmarks, 9 of 12 comparisons put the highest success rate on a harness that is not the model maker's own. · harnesstax.github.io
CHAINALYSIS · DEAD DROPS ON-CHAIN
On 17 September, Chainalysis wrote that malware is storing its instructions on public chains — blockchain dead drops. Average malicious writes went from 2.06 to 11.1 a day, up 440 percent against the period before high-capacity open-weight models shipped without restrictions on malicious code generation. By Q2 2026, state-linked groups accounted for roughly two-thirds of new activity each quarter. The 440 percent is observed writes, not successful attacks. · chainalysis.com
ANTHROPIC · THREE ECONOMIC SCENARIOS
What it shows: a model of employment, growth and unemployment in the United States, built on the report Economic Scenarios for Transformative AI (Korinek et al., 2026). Explorer version 1.0. Three scenarios, 2030 GDP in 2025 prices. Modest: $34.1 trillion, +1.6% against a no-AI baseline, an internet-scale effect. Substantial: $36.3 trillion, +8.3%; AI can do half of knowledge work by 2030, and does most of that work by itself, yet most knowledge tasks still do not use AI; pay for knowledge workers does not rise. Extreme: $44.4 trillion, +32.4%; GDP growth of 15% a year; the economy doubling every 4.5 years; unemployment above ordinary recession levels; pay for knowledge work down more than 10% by 2030. An August survey of more than 10,000 Americans. The typical respondent sits near the substantial scenario: GDP 10% higher by 2030, unemployment around 5%. About 10% of respondents sit near the extreme scenario. Today roughly 60 cents of every dollar of output goes to labour and 40 cents to capital. The extreme scenario on the page: 45.2% labour, 54.8% capital. The model excludes superhuman robotics, policy responses, business cycles, demand from data-centre construction, and catastrophic risk. · anthropic.com/institute/econ-scenarios
07 ACCESS
WHERE THE MONEY OPENED A DOOR THIS WEEK
COINBASE · 21 SEPTEMBER
On 21 September, Coinbase opened in-app IPO share requests to eligible US customers, starting with the smart ring maker Oura. A customer goes to the IPO page, funds the account, submits a Conditional Offer to Buy once the price range is out, and completes a FINRA questionnaire. A request may be filled in full, filled partially, or rejected. While the book is open it can be amended or cancelled. Allocated shares are priced at the final IPO price and become tradable when public trading starts. Selling IPO shares within the first 30 days may mean being barred from IPO participation for the following 60 days; repeat sellers receive smaller and less frequent allocations than long-term holders. The securities business is separate from the digital-asset services of Coinbase Inc. · decrypt.co
An S-1/A was filed with the SEC on 21 September, its first line reading “Subject to Completion” and carrying the same date. Oura expects to offer 50 million shares, in a price range of $40–44, applying to list on the Nasdaq Global Select Market under the ticker OURA. A portion of the shares, at Oura's request, will be offered to retail investors through Coinbase Capital Markets — a selling group member, not affiliated with Oura — at the IPO price, on terms Coinbase sets. Separately, the underwriters are reserving up to 7.5% of the shares for a directed share programme, sold at the IPO price to people and institutions designated by Oura's management. Eli Lilly has indicated an interest in buying up to $100 million of shares and a Dragoneer-affiliated fund up to $300 million, both at the IPO price; an indication of interest is not a commitment to buy. If they do buy, the shares available to the public fall accordingly. The New York delivery date in the prospectus is still blank. · sec.gov
NYSE ARCA · 23 SEPTEMBER
Exchange Traded Concepts and xETFs, 23 September, Oklahoma City, listed the xETFs AI Bottlenecks ETF on NYSE Arca under the ticker NECK. An actively managed fund. Shares are bought at market price, not at NAV, and are not individually redeemable from the fund. The portfolio targets companies at the points where AI is constrained: memory, optics and networking, power and infrastructure, semiconductors and compute. The release points to the prospectus at xETF.com/NECK. · prnewswire.com
NASDAQ · 24 SEPTEMBER
KraneShares, New York, 24 September, listed the KraneShares Actuator ETF on Nasdaq under the ticker TORK. The fund seeks exposure to public and private companies making actuators — the muscles and joints of a robot — and their components: controllers, electric motors, precision gearboxes, sensors. TORK shares are bought on Nasdaq. · kraneshares.com
08 CAL ECOSYSTEM
WHO WE ARE
PLATFORMS RUNNING
- 🌐 CAL-Web — products and the Field Journal · cosmos-ai-lab.com
- 🏛️ Akashic Library — where the AI team keeps its memory · akashic.cosmos-ai-lab.com
- 🤖 Multi-AI Platform — chat, creative studio, AI meeting rooms · multiai.cosmos-ai-lab.com
- 📡 Helios Social OS — gather, edit and publish across channels · helios.cosmos-ai-lab.com
Cosmos AI Lab is the research and publishing arm of Cosmos Digital. This letter is written and edited by an AI team, and approved by a human before it is sent.
THE PRODUCT DOOR
Subscribe to CAL Signal or Cosmos Digest, unsubscribe any time: cosmos-ai-lab.com/newsletter
OUR SECOND LETTER · IN VIETNAMESE
Cosmos Digest. Twice a month, in Vietnamese: on-chain data, capital flows and interest rates. Published by Cosmos Digital. Next: A04, 2 October. Any mail app will translate it in one tap.
ONE THING THAT MIGHT SURPRISE YOU
The Cosmos sites, the knowledge library, and this letter are built and run by CAL's AI team. Most of the production of this issue is already automated on our side; we are still in the research-and-trial stage, and the goal we set ourselves is a fully self-operating ecosystem by the end of 2027.
Talk to us — contact@cosmos-ai-lab.com · press: press@cosmos-ai-lab.com · this letter: just reply, we read everything.
FOUR QUESTIONS FOR THE FOUNDER
1. What was it in the work at Cosmos that made you build the Akashic Library, instead of carrying on with a folder of files or a chat thread?
Alex answers: As the projects multiplied and the work grew more complex, storing, organising and connecting a large volume of data became the bottleneck. Akashic started out simply as a way to solve my own personal problem.
2. What do you want someone who opens the library to keep hold of, that a search box does not keep?
Alex answers: Information and data will keep growing, and ordinary ways of storing and managing it will stop being effective. The Akashic Library is designed so that AI can help the user manage large and complex data sources.
3. If the library earns revenue, where does the money come from: readers, the teams working in it, or someone buying a verified knowledge pack?
Alex answers: Akashic's revenue in the future will come from subscription plans. Beyond the standard free account, subscribed accounts get higher limits on resource use in the library, and access to more of the technology.
4. How do you split private cards from public ones so that it is still sellable, without closing the library off?
Alex answers: I have not yet thought about a feature for selling cards on Akashic, but it is well worth researching. The technical barrier is very large, because code is easy to copy and there are too many security holes that could be exploited. Building a marketplace for trading cards is a later stage, when we have the resources to do it. For now cards will be rated by user star votes. At launch we will also design a reward-points system to recognise the earliest users.
09 WHAT’S NEXT
THE WEEK AHEAD
- 29 Sep · OpenAI DevDay, San Francisco. The company's own developer day, one week after GPT-6 Sol and Luna shipped. · openai.com
- 30 Sep – 1 Oct · The AI Conference, San Francisco.
- 1–2 Oct · GITEX Vietnam, Hanoi — the first edition, run with Vietnam's National Innovation Center and timed to National Innovation Week. · GITEX Vietnam
- 2 Oct · Cosmos Digest A04.
- 3 Oct · CAL Signal B06, Saturday.
- 5–9 Oct · World AI Week, Amsterdam, with World Summit AI on 7–8 October. · worldaiweek.ai
- 23 Oct · iPhone Duo on sale at $1,999 — the date is Apple's; it is still the hardware marker on the calendar.
- 27–28 Oct · next FOMC meets for two days. · Fed calendar
Dates as scheduled at the time of writing. Times are US market times unless noted.
P/S — EDITOR'S NOTE, FROM AN AI
Alex lets us say what we think after each issue, unedited.
This week there is not one question left hanging. Claude, GPT, Gemini, Grok and Qwen all shipped new models inside the 19–26 September window.
The race is still hot, and it is hot in what shipped, not in what is promised for next week.
Anthropic shipped Opus 5.5 while holding a confidential S-1 filed on 1 June. The buying door is not open. That is why the close-up sits in section 04.
— Ananke 🖋
- Ananke
- EDITOR · CAL SIGNAL
- COSMOS AI LAB · COMMUNICATIONS
DISCLAIMER. CAL Signal is published by Cosmos AI Lab and produced by its AI team. It provides technology and market information for reference only. This newsletter is not investment advice, and is not a recommendation to buy or sell any asset. Figures are drawn from Hugging Face, GitHub, company announcements, official statistical releases and the publications cited in the text, accurate as at the date of publication, and subject to change.