Lookonchain APP

App Store

First official deployment of a U.S.-developed open-source model: Inkling 2 Small has been open-sourced, requiring at least 500GB of VRAM to run.

25 minutes ago

Beating AI Express News: Thinking Machines, founded by former OpenAI CTO Mira Murati, has officially opened the weights of its Inkling 2 Small model. The model is now available on Hugging Face under the Apache 2.0 license, allowing download, fine-tuning, and commercial use. Inkling 2 Small launched last week, with a composite score of 31 from Artificial Analysis, making it the highest-scoring open-source model outside China to date. It features 594 billion total parameters, with 17 billion activated per inference, supports image input, and offers a maximum context window of 1 million tokens. Despite being labeled "Small", it has a fairly high operational threshold. The official provides two weight versions: the NVFP4 variant requires at least 500GB of GPU VRAM, equivalent to 2 NVIDIA B300s or 4 H200s; the MXFP8 version needs a minimum of 800GB VRAM, with evaluation scores only 1 to 2 percentage points higher than the former. Currently, the model supports inference frameworks including vLLM, SGLang, and Unsloth, and can also be accessed via APIs on platforms such as NVIDIA, Fireworks, Together, and Cloudflare.

Relevant content

Bank of America: Global equity funds see record annualized inflows of $1.4 trillion; investor sentiment is bearish but short positions remain scarce.

Strategists at Bank of America Corp. said global equity funds are attracting record inflows this year, with few investors willing to take short positions. The team led by Michael Hartnett wrote in a report that equity exposure has surged sharply, with an annualized $1.4 trillion flowing into global funds. Of that total, U.S. equity funds drew $656 billion, tech stocks drew $222 billion, and South Korean equities drew $78 billion. They noted: "These inflows provide impressive support for the stock market, but also show that while many investors are sentimentally bearish, few are actually putting their money into short positions or underweighting assets." Hartnett added that risk assets have struggled over the past six months "because bullish positions collided with policy tightening." Still, the strategists maintain a "risk-off stance until interest rate hikes end and financial conditions ease." They believe this shift will only occur when U.S. crude oil prices fall below $70 per barrel, the U.S. dollar reverses its recent rally, and the 10-year U.S. Treasury yield drops below 4.5%.

4 minutes ago

Telegram Founder on Jensen Huang's Departure: Founders Should Take Control of Their Own Boards

Telegram founder Pavel Durov commented on Jensen Huang's departure, noting the "rookie mistake" Huang made was establishing a board of directors. Durov recalled he was ousted by his board in 2014 after refusing demands from intelligence officials. Amid the intelligence allegations surrounding Huang, Durov quipped that whether a founder says "yes" or "no" to authorities, a board could still oust them. He ended with advice for entrepreneurs: "Don't set up a board of directors—be your own board."

4 minutes ago

OpenAI and NVIDIA are purchasing its data: Databento, a firm with only 39 employees, has become a unicorn.

Beating AI Express News reports that financial data firm Databento has closed a $100 million Series C funding round, pushing its valuation to $1.1 billion. The Series C comes just four months after it secured a $97 million Series B round in July this year. The company currently has only 39 employees and has already been profitable for some time. According to official disclosures, its revenue has more than doubled since the previous funding round. The company had no plans to raise capital this time, but existing shareholders came forward to request additional investment. Databento operates in the financial market data space. It sources real-time and historical trading data for stocks, futures, and options directly from exchanges, then delivers this data via APIs to quantitative funds, brokerages, and developers. Clients can purchase data on-demand, allowing their programs to access the data directly without going through the complex procurement processes associated with traditional financial data services. AI firms have also emerged as key clients for Databento. The company says five of the world’s six leading AI labs have signed enterprise contracts with it, with OpenAI and NVIDIA among its customers. Founder Christina Qi has disclosed that roughly 90% of the company’s revenue comes from large financial institutions and AI labs. The company plans to use the new funds to expand its global data infrastructure. Over the past four months, it has advanced the deployment of nine data centers in locations including Hong Kong, Japan, and London, with plans to offer historical trading data from additional markets.

4 minutes ago

a16z publishes a comprehensive landscape of the Agent Economy, with crypto-related projects such as Kite mostly concentrated in payment and wallet infrastructure.

Venture capital firm a16z recently released a comprehensive overview of the agent economy. In a companion article, a16z partners Olivia Moore and Justine Moore noted that as AI agents begin to select products, subscription services, and complete purchases on behalf of users, payment, identity verification, and anti-fraud infrastructure will face new demands. The "Payments & Checkout" section of the overview includes crypto-related projects such as Kite, Crossmint, Privy, Turnkey, and Catena Labs, covering stablecoin payments, wallets, and expense management. Among these projects, Kite is the only one that has issued a tradable native token to date.

4 minutes ago

Analyst: Bitcoin’s momentum indicators have weakened; if historical bearish patterns repeat, BTC could drop below $60,000.

Crypto analyst Omkar Godbole noted in a post that Bitcoin’s weekly MACD histogram dropped below the zero line this week, signaling weakening price momentum. The weekly bearish signal will only be confirmed if the histogram remains negative at Sunday’s weekly close. At press time, Bitcoin trades at around $69,000; a 16% drop would push it below $60,000, approaching this year’s low of roughly $57,700. Statistics show that since November 2012, Bitcoin’s weekly MACD has generated 24 bearish signals. Of those, in 13 cases, Bitcoin fell at least 16% from the signal confirmation before the indicator turned bullish again. Among these 13 instances, 5 saw the decline hit within a week, 10 within four weeks, with a median time of three weeks. Notably, in November 2018 and February 2014, Bitcoin dropped 38% and 35% respectively within a week after signal confirmation. Conversely, the other 11 times saw declines of less than 16%. The most recent bearish signal occurred in June this year, when Bitcoin only fell around 3% to $57,700 before rebounding. The current signal emerges amid outflows from U.S. spot Bitcoin ETFs and a strengthening U.S. dollar, though a single technical indicator is insufficient to determine whether Bitcoin will retest this year’s low.

4 minutes ago

US Department of Justice seizes cloud infrastructure used in North Korea-linked 'FaceStealer' attack, tied to over $400 million in stolen cryptocurrency.

According to a U.S. Department of Justice (DOJ) announcement, U.S. law enforcement agencies on local time November 12 unveiled two actions targeting North Korea-linked cryptocurrency theft operations: seizing cloud infrastructure used in the "FaceStealer" attack campaign, and in a separate Binance theft case, U.S. government and private partners have collectively frozen over $2.7 billion in involved assets, with more than $1.2 billion authorized by courts for seizure or transfer restrictions. FaceStealer has caused over $400 million in losses over the past 18 months. The FBI said the group uses video conferencing and deepfake videos to impersonate contacts trusted by targets or prominent crypto industry figures, luring victims to install malware. Beyond crypto holders and Web3 workers, attackers are increasingly targeting open-source software developers and maintainers to gain access to their devices and code repositories. Acting on federal search warrants, the FBI seized more than 900 victim camera recordings and issued prevention and response guidelines. On the Binance theft case: The over $1.2 billion forms part of the frozen assets. The DOJ stated that around October 29, the North Korean military-linked hacker group dubbed "TraderTraitor" by the FBI stole over $6 billion in crypto assets from Binance. Following the incident, law enforcement agencies including the DOJ and FBI collaborated with virtual asset service providers to freeze a portion of the stolen funds. The DOJ added that tracking, freezing, seizure and forfeiture procedures are ongoing, and the frozen amount does not equal the total to be returned to victims.

4 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano