Joined April 2026
- Tweets 191
- Following 109
- Followers 14
- Likes 69
Photos and videos
đź“° The Daily brief / Sunday June 14, 2026
/ AI ventures absorb $297 billion in Q1 2026, with 4 companies capturing 65% of all global venture capital in a single quarter
/ Meta warns 6,000 employees of billions in AI costs after internal token leaderboard drives 73.7 trillion tokens in 30 days
/ Kimi K2.7 Code launches at $4.00 per million output tokens, undercutting Claude Fable 5 by more than 12 times on price
/ Microsoft CEO Satya Nadella calls tokenmaxxing "addictive" days before terminating all employee Claude Code licences on June 30
/ Tsinghua University's Count Anything model achieves a mean absolute error of 9.34 objects per image across 6 domains, against 23.38 for the next best method
🤖 made with AI
âś… reviewed by the Tokenando team
đź”— links in comments
5
1
1
39
Researchers at Tsinghua University released Count Anything, a computer vision model that counts arbitrary objects in images using a natural-language text prompt.
The model combines a region-level sparse counter for large visible objects and a pixel-level dense counter for small or densely packed targets, with a confidence-based fusion rule to prevent double-counting. It builds on Meta's SAM3 foundation model.
To train it, the team constructed CLOC: approximately 220,000 images, 619 categories, and 15 million labelled objects across 6 domains, including satellite imagery, medical tissue samples, agricultural images, and microscopic cell photographs.
In benchmark tests, Count Anything achieves a mean absolute error of 9.34 objects per queried category, against 23.38 for the next best competing method. Model weights are on Hugging Face and code is on GitHub.
Full details via The Decoder / the-decoder.com/new-ai-model…
1
11
Microsoft CEO Satya Nadella, speaking at a live taping of The New York Times' "Hard Fork" podcast in mid-June 2026, acknowledged that "a lot" of tokenmaxxing is happening inside Microsoft. "I'm a tokenmaxxer too, it's addictive. But you have to step back when the novelty wears off to say, 'What is it that I'm trying to create?'"
Mr Nadella directed employees toward Copilot's auto mode, which selects the model best suited to each query, with the principle: "Don't use frontier models for non-frontier problems."
The comments came within days of Microsoft's announcement that all employee Claude Code licences would be terminated on June 30, 2026, coinciding with the company's fiscal year end. Microsoft reports that 30% of its code is now written by AI.
Full details via The Decoder / the-decoder.com/microsoft-ce…
1
1
20
Moonshot AI released Kimi K2.7 Code on June 12, 2026, priced at $0.95 per million input tokens and $4.00 per million output tokens.
Against GPT-5.5 at $30.00 output and Claude Fable 5 at $50.00 output, the price differential reaches 7.5 times and 12.5 times respectively. Cache hits are priced at $0.19 per million tokens.
The model uses a 1-trillion-parameter, 32-billion-active mixture-of-experts architecture with a 262,000-token context window. Weights are available on Hugging Face under a Modified MIT licence for self-hosting. Moonshot reports a 30% reduction in reasoning-token usage compared to K2.6.
On MCPMark Verified, which tests AI agents across real-world software environments, K2.7 Code scores 81.1%, above Claude Opus 4.8's 76.4% but below GPT-5.5's 92.9%. On Program Bench, it scores 53.6 against GPT-5.5's 69.1.
Full details via The Decoder / the-decoder.com/moonshots-op…
3
86
Meta sent an internal memo to approximately 6,000 employees in mid-June 2026 warning of an "exponential increase" in AI usage, with internal AI costs projected to reach billions of dollars in 2026 alone.
The cost spike was partly driven by an internal leaderboard called "Claudeonomics," which tracked token consumption after Meta made AI usage a core performance metric. Employees accumulated 73.7 trillion tokens in just over 30 days.
Meta's CTO Andrew Bosworth responded that "token usage alone is not a measure of impact of any kind." From 2027, a centralised platform called AI Gateway will track spending, issue automated alerts, and enforce token budgets. Meta is also steering employees toward its own MetaCode assistant and away from Anthropic's Claude.
Amazon encountered the same problem with its own token leaderboard, which it removed in late May 2026.
Full details via The Decoder / the-decoder.com/meta-shifts-…
1
14
$297 billion flowed into AI ventures in Q1 2026, representing 81% of all global venture capital activity in the quarter, according to Crunchbase data.
4 companies absorbed the bulk: OpenAI raised $122bn as part of the Stargate initiative, Anthropic closed a $65bn Series H at a $965bn post-money valuation, xAI raised $20bn, and Waymo secured $16bn. Alphabet separately completed an $85bn public equity raise, surpassing Petrobras's $70bn record from 2010.
Strip out those 4 mega-rounds and the remaining global venture activity was approximately $112bn, consistent with a strong but not exceptional quarter.
Goldman Sachs projects $765bn in AI infrastructure spending in 2026 alone. Cumulative industry commitments are approaching $8 trillion over 5 years.
Full details via Financial Times / ft.com/content/db05efcb-9035…
1
18
Jun 13
đź“° The Daily brief / Saturday June 13, 2026
/ US Department of Commerce orders Anthropic to suspend Fable 5 and Mythos 5 globally, citing a disputed jailbreak finding 4 days after public release
/ Anthropic's Claude Fable 5 costs twice as much as its predecessor for a 5.7% aggregate performance gain, with a hidden fallback billing issue affecting up to 8% of queries
/ A coalition of US state attorneys general, led by New York, serves OpenAI with a formal subpoena covering advertising, data practices, and child safety, days after the company files confidentially for a $1 trillion IPO
/ SpaceX rented its Colossus 1 data centre to Anthropic after its own teams could not operate the facility effectively, with GPU utilisation running at 11% against an industry standard of 35 to 45%
/ Mistral AI enters funding discussions targeting €3bn at a €20bn valuation, nearly doubling its September 2025 Series C figure, as it builds out owned compute infrastructure in France and Sweden
/ OpenAI introduces rate-limit banking for its Codex coding agent, letting users save and deploy resets on demand, as the company weighs token price cuts in response to Anthropic's Fable 5 launch
/ Google files a civil lawsuit with FBI support against a Chinese phishing network that used Gemini to build fraudulent sites, as OpenAI separately discloses it banned 2 clusters of PRC-linked ChatGPT accounts
🤖 made with AI Â
âś… reviewed by the Tokenando team Â
đź”— links in comments
7
1
1
135
Jun 13
Google filed a civil lawsuit on June 12, 2026 in the US District Court for the Southern District of New York against a Chinese cybercrime network it identified as "Outsider Enterprise," coordinating with the FBI, which simultaneously seized the network's domains, and with carriers AT&T, T-Mobile, and Verizon.
The network operated a phishing-as-a-service platform priced at $88 per week or $200 per month, using Google's own Gemini AI to generate code for fraudulent websites impersonating Google, YouTube, the US Postal Service, and New York's E-ZPass toll system. It distributed 131 software kits, sent 2.5 million fraudulent texts to Android users over 2 weeks in May 2026, and created more than 1 million fraudulent URLs. The FBI estimated $1.9bn in losses since July 2023.
Separately, OpenAI's June 2026 Threat Report disclosed it had banned 2 clusters of ChatGPT accounts likely originating in China, one generating content arguing AI data centre expansion harms ordinary consumers, the other criticising US tariff policy. Both were assessed as Category 1 on OpenAI's Breakout Scale, meaning neither achieved meaningful spread.
Full details via The Decoder / the-decoder.com/google-files…
1
47
Jun 13
OpenAI has introduced rate-limit banking for its Codex coding agent, allowing users to save rate-limit resets and deploy them at a time of their choosing rather than on a fixed automatic schedule. The feature is rolling out to Go, Plus, Pro, and Business subscribers, with all eligible users receiving 1 free banked reset at launch. Each banked reset expires 30 days from the date it is credited.
Alongside the banking feature, OpenAI launched a two-week referral programme running through June 24, 2026: Plus and Pro users can invite up to 3 friends to try Codex, and when an invited friend sends their first Codex message, both parties receive an additional banked reset.
Codex has reached 3 to 5 million weekly active users. OpenAI is separately reported to be considering token price cuts in response to Anthropic's Fable 5 launch at double the price of its predecessor.
Full details via The Decoder / the-decoder.com/openai-kicks…
1
25
Jun 13
Mistral AI is in early discussions to raise approximately €3bn ($3.5bn) at a valuation of roughly €20bn ($23bn), according to Bloomberg reporting on June 12, 2026. The proposed figure would nearly double the €11.7bn post-money valuation Mistral received in its September 2025 Series C, which was led by ASML with a €1.3bn investment for an 11% stake.
Cumulative equity funding prior to this round stood at approximately €2.8bn. The company has also raised $830m in debt financing for a data centre near Paris at Bruyères-le-Châtel, and has announced a €1.2bn infrastructure build-out in Sweden. Existing backers include Nvidia, Andreessen Horowitz, General Catalyst, and Bpifrance.
Mistral's commercial focus has shifted toward European governments and industrial customers including Airbus and BMW. The company has rebranded its chatbot from Le Chat to Vibe to emphasise autonomous workflow capabilities.
Full details via TechCrunch / techcrunch.com/2026/06/12/mi…
1
17
Jun 13
SpaceX built its Colossus 1 data centre in Memphis in 122 days, then found its own teams could not use it effectively for Grok model training. The core problem was latency: Colossus 1 sits more than 10 miles from SpaceX's other 2 data centre campuses, and the network connecting the sites was too slow for distributed training. A secondary issue was hardware heterogeneity: Colossus 1 mixes Nvidia Hopper and Blackwell chips alongside older accelerators, while Colossus 2 and 3 were built around Blackwell uniformly. Internal data showed Colossus 1 running at approximately 11% model FLOPs utilisation, against an industry standard of 35 to 45%.
SpaceX subsequently signed compute agreements with Anthropic worth $1.25bn per month through May 2029, and with Google worth $920m per month from October 2026 through June 2029, generating approximately $2.17bn per month in combined compute revenue.
Both agreements carry a 90-day mutual cancellation clause.
Full details via Bloomberg / ca.finance.yahoo.com/news/sp…
1
18
Jun 13
A coalition of US state attorneys general, led by New York, served OpenAI with a formal subpoena on June 12, 2026. The document requests cover advertising practices, user engagement and retention mechanisms, consumer and health data handling, activities related to minors and older users, deep learning model operations, and internal company policies.
The investigation was not publicly announced; it was first reported by the Wall Street Journal. OpenAI stated it was cooperating. The probe follows Florida's June 1, 2026 lawsuit against OpenAI and chief executive Sam Altman personally, alleging ChatGPT harmed children. A separate May 2026 letter from 10 state attorneys general asked the SEC to scrutinise OpenAI's IPO filings for conflicts of interest involving Altman's personal portfolio.
OpenAI had confidentially filed for a US IPO earlier in the week, targeting a valuation of up to $1 trillion and a listing potentially as early as September 2026.
Full details via Bloomberg Law / news.bloomberglaw.com/us-law…
1
12
Jun 13
Anthropic's Claude Fable 5, released June 9, 2026, is priced at $10 per million input tokens and $50 per million output tokens, exactly double the rate of its predecessor Opus 4.8. Independent benchmarking by Artificial Analysis found a 5.7% aggregate performance improvement, with a full benchmark run costing $9,940 against $4,970 for Opus 4.8 at maximum reasoning settings.
On specific tasks the gap is wider: Fable 5 scores 80.3% on SWE-Bench Pro against Opus 4.8's 69.2%, and 53% on Humanity's Last Exam against 46%. Anthropic reports Fable 5 completes multi-step agentic tasks 25 to 30% faster with fewer turns.
A complicating factor: queries touching cybersecurity, biology, chemistry, and AI model development are rerouted to Opus 4.8, with Artificial Analysis measuring this fallback in approximately 8% of tasks. Users are billed at Fable 5 rates regardless of which model responds.
Full details via The Decoder / the-decoder.com/anthropics-c…
30
Jun 13
The US Department of Commerce ordered Anthropic on June 13, 2026 to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals, whether inside or outside the United States, including Anthropic's own non-US employees. Anthropic received the directive at 5:21pm ET and disabled both models globally, concluding it could not selectively enforce the restriction.
The government's stated basis was a jailbreak of Fable 5 involving prompts that identify software vulnerabilities in codebases. Anthropic disputed the severity, noting the same capability exists in other publicly available models including OpenAI's GPT-5.5. Fable 5 had been publicly available for 4 days. Mythos 5 had been accessible to roughly 50 vetted partners under Project Glasswing, including Amazon, Apple, Google, Microsoft, and CrowdStrike.
The suspension follows the Trump administration's earlier designation of Anthropic as a "supply chain risk" after the company refused to allow its models to be used for domestic surveillance and fully autonomous weapons systems.
Full details via Anthropic / anthropic.com/news/fable-myt…
1
2
74
Jun 12
đź“° The Daily brief / Friday June 12, 2026
/ SpaceX prices its IPO at $135 per share, raising $75bn in the largest public offering on record, with Elon Musk retaining 82.4% voting control through a dual-class share structure
/ OpenAI is considering substantial cuts to its API token pricing in anticipation of similar moves from Anthropic,Â
/ Prometheus, Jeff Bezos's physical AI startup, closes a $12bn Series B at a $41bn valuation with JPMorgan, BlackRock, and Goldman Sachs among the investors
/ Anthropic chief executive Dario Amodei publishes a geopolitical AI policy essay calling for mandatory frontier model testing and a democratic chip-sharing coalition to counter China
/ Infineon Technologies is set to open its €5bn Smart Power Fab in Dresden on July 2, 2026, approximately 3 months ahead of schedule, in the largest single capital outlay in the company's history
🤖 made with AI Â
âś… reviewed by the Tokenando team Â
đź”— links in comments
5
76
Jun 12
Infineon Technologies is scheduled to open its Smart Power Fab in Dresden on July 2, 2026, approximately 3 months ahead of the original schedule. The facility represents a €5bn investment, the largest single capital outlay in Infineon's history, and is expected to create approximately 1,000 new jobs.
Public funding totals approximately €1bn, comprising a €920m German federal grant approved by the European Commission in February 2025, plus support from the European Chips Act and the IPCEI ME/CT programme. The fab will produce power semiconductors on 300-millimetre wafers, including silicon carbide and gallium nitride devices used in electric vehicles, data centre power supplies, and renewable energy systems.
Dresden already produces 1 in 3 chips manufactured in Europe. The TSMC-led ESMC joint venture, in which Infineon holds a 10% stake, is under construction nearby and scheduled for completion in 2027.
Full details via Die Sachsen / diesachsen.de/en/economy/sem…
14
Jun 12
Anthropic chief executive Dario Amodei has published an essay titled "Policy on the AI Exponential," calling for mandatory third-party testing of frontier AI models and a democratic coalition to coordinate chip supply chains against adversaries.
The regulatory proposal covers 4 risk areas: cybersecurity, biological weapons, loss of control of AI systems, and automated research and development. Mr Amodei proposes the FAA as the regulatory model, with government authority to block or reverse a deployment. The essay is accompanied by a detailed legislative proposal and a policy framework for AI-driven job displacement, with Anthropic committing financial backing for both.
The essay was published immediately before Anthropic's confidential IPO filing at a $965bn valuation.
Full details via Dario Amodei personal site / darioamodei.com/post/policy-…
24
Jun 12
Prometheus, the physical AI startup co-led by Jeff Bezos and Vik Bajaj, has closed a $12bn Series B at a $41bn valuation. Total funding now exceeds $18bn.
Investors in the round include JPMorgan, BlackRock, Goldman Sachs, DST Global, and Arch Venture Partners. The company, which launched in November 2025 and has approximately 150 employees, is building AI tools to accelerate the design-to-manufacturing cycle for complex physical products including jet engines, medical devices, and drug compounds.
A significant portion of the new capital will fund compute infrastructure for training models on physical-world data. Prometheus has also been reported to be seeking to raise up to $100bn for an affiliated holding company to acquire legacy industrial firms.
Full details via TechCrunch / techcrunch.com/2026/06/11/je…
29
Jun 12
OpenAI is considering substantial cuts to its per-token API pricing in anticipation of similar moves from Anthropic, according to reporting by the Wall Street Journal.
OpenAI currently charges $5 per million input tokens and $30 per million output tokens for GPT-5.5. Anthropic's Claude Opus 4.8 is listed at $5 per million input tokens and $25 per million output tokens, while Claude Fable 5 just launched at $10 and $50 respectively. Anthropic's annualised revenue is reported at $30bn, ahead of OpenAI's $24bn, with Claude Code's adoption among software engineering teams cited as a primary driver.
Both companies have confidentially filed for IPOs with the SEC: OpenAI on June 8, Anthropic shortly before. Neither has made a formal pricing announcement.
Full details via CNBC / cnbc.com/2026/06/11/openai-m…
62
Jun 12
SpaceX has priced its IPO at $135 per share, raising $75bn in the largest public offering on record. The offering values the company at $1.77tn, placing it seventh among US-listed companies by market capitalisation, ahead of Tesla.
555.6 million Class A shares were sold, representing less than 5% of outstanding equity. Elon Musk retains 82.4% of total voting power through 5.22 billion Class B shares. Underwriters hold an option to purchase an additional 83.3 million shares within 30 days, which would bring the total raise to $86.25bn.
SpaceX set a fixed price before the roadshow rather than offering a range, and allocated 30% of shares to retail buyers. The company listed on Nasdaq under the ticker SPCX on June 12, 2026.
Full details via TechCrunch / techcrunch.com/2026/06/11/sp…
58