Lookonchain APP

App Store

Opus 5.5 has become the first Opus iteration to restrict cutting-edge large language model development: Once triggered, it downgrades to Opus 5.

33 minutes ago

Beating AI News: Anthropic has for the first time integrated its "cutting-edge model development" restrictions from the Fable series into its Opus line. Opus 5.5 now includes a Fable-like classifier. When handling a small number of cutting-edge LLM development tasks—such as developing kernels for specific AI accelerators—the system automatically switches from Opus 5.5 to the less powerful Opus 5, which does not have this restriction. Fable 5 had already restricted tasks including distributed training infrastructure development, AI accelerator design, and certain kernel development. At the time, Anthropic explained that powerful models could accelerate next-generation model development, and it did not want Claude to help competitors build cutting-edge models of the same level faster. Fable currently also blocks or switches models for a small subset of such tasks. Anthropic’s internal evaluations also have these safeguards enabled. Once a cutting-edge LLM development task triggers the restriction, Opus 5 takes over, which may lower Opus 5.5’s related benchmark scores. Anthropic emphasized that this type of classifier covers only a very small portion of cutting-edge LLM development tasks, and most general AI, machine learning research, and daily programming will not be affected. After a downgrade is triggered, Claude will clearly indicate that it has switched to Opus 5, and subsequent conversations will continue using Opus 5.

Relevant content

Former partner at Hack VC reported dead; he alleged before his death that the company forced him to work while ill.

According to crypto industry practitioner Jordy’s disclosure and local police reports, former Hack VC partner and platform head Hsin-Ju was found dead near the on-ramp of Interstate 15 in California’s Mojave Desert on August 24, at the age of 37. Her death occurred exactly one day after she announced plans to expose Hack VC. Investigating authorities have not yet determined the cause of death, and it remains unclear whether heat, sudden illness, mechanical failure, or another factor was responsible. Earlier, on August 23 local time, Hsin-Ju posted on Twitter that she had rejected Hack VC’s settlement offer and would release evidence related to her tenure at the firm on August 26. She had previously alleged that during her time at Hack VC, she was ordered to continue working despite a medical emergency, and was threatened by the company that if she resigned early before completing an event, she would be blacklisted in the industry. Following her departure, she accused the firm of mishandling her health insurance and continuing to pressure her during legal proceedings. She has since fired her lawyer and stated she would rather forgo a settlement than keep the matter private. In response, Hack VC posted on Twitter that it had seen Hsin-Ju’s tweet and was deeply concerned about her well-being. The firm added that its understanding of the matter differs significantly from Hsin-Ju’s claims, but out of respect for her privacy, it does not wish to discuss details publicly at this time.

2 minutes ago

Bitcoin falls below $86,000, with its 24-hour price gain narrowing to 0.96%

According to HTX market data, Bitcoin has dropped below $86,000, with its 24-hour price gain narrowing to 0.96%.

2 minutes ago

TinyFish launches AI monitoring tool Monitor, eliminating the need for agents to repeatedly scrape web pages.

Insight Beating AI News Flash: Web Agent firm TinyFish has launched Monitor. Users input a webpage or topic, then specify their area of concern in one sentence. Monitor runs continuous checks, notifying users or agents only when a qualifying change occurs. For example, if you set it to track a product page, it will alert you exclusively when the price drops below $500. The tool regularly crawls webpages and compares results from prior scans. Upon detecting a change, it uses AI to confirm if the change aligns with your requirements—routine text adjustments will not trigger alerts. Monitor supports both fixed webpage monitoring and continuous topic searching, enabling tracking of changes including prices, inventory, job postings, new products, and brand mentions. Alerts are delivered via email and Webhook. Currently, each account can run up to 50 Monitors simultaneously, with checks as frequent as every 30 minutes. The service is free until September 25.

2 minutes ago

Genius. fun now supports deploying trading pairs combining tokens and AMCB.

On-chain trading terminal Genius Terminal, which is invested by YZi Labs and has CZ as its advisor, announced that its token launch platform Genius.fun now supports pairing deployed tokens with AMCB. AMCB is the Binance bStocks version of the tokenized U.S. cinema chain AMC Entertainment (ticker: AMC). BlockBeats Note: Stock Meme is an emerging concept that directly combines traditional meme coins with tokenized U.S. stocks. Instead of pairing with USDT or ETH, meme coins are now paired directly with on-chain U.S. stock tokens (such as NVDA, TSLA, AAPL, etc.). This model retains the high volatility and community-driven speculative attributes of meme coins, while leveraging the popularity and narrative of real stocks. Typically, a portion of transaction fees flows back to the community treasury to accumulate the corresponding U.S. stock tokens, forming a dual-driven model of "sentiment-driven speculation + real asset anchoring".

2 minutes ago

Qwen releases Qwen-Audio-3.1, cuts prices across the entire Qwen-Audio line of speech models.

Beating AI News: Qwen has officially launched the Qwen-Audio 3.1 series of large speech models. This upgrade not only brings comprehensive improvements to three core models—Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and real-time speech interaction (Realtime)—but also introduces two new key models: the audio creation model Qwen-Audio 3.1-TTS-Next and the audio understanding model Qwen-Audio 3.1-ASR-Next. The five new speech models launched together form a complete audio capability stack covering "understanding, generation, interaction, and creation". To further lower user costs, all Qwen-Audio series speech models have had their prices reduced: TTS by approximately 70%, Realtime by around 85%, and ASR by up to 95%.

2 minutes ago

Binance to list OURA U.S. dollar-margined U.S. stock Pre-IPO perpetual contracts

Binance announced it will launch U.S. dollar-margined U.S. stock Pre-IPO perpetual contracts for OURA at 16:30 (UTC+8) today, with the underlying asset being U.S. stock Oura Inc. and supporting up to 20x leverage.

2 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano