Arthur Hayes: I bet that Bitcoin will surpass $110,000 before retracing to $76,500

According to monitoring by AI Aunt, a whale address that opened a $10 million altcoin position starting September 23 withdrew 2.455 million ONDO from Binance 12 hours ago, valued at approximately $1.144 million with an average withdrawal price of $0.4662. The address currently holds around $9.218 million in three altcoins on-chain: 1.014 million UNI (worth ~$8.91 million, average withdrawal price $8.79, unrealized loss ~$1.551 million), 2.455 million ONDO (valued at ~$1.144 million), and 12,400 LTC (worth ~$841,000, average withdrawal price $67.86, unrealized loss ~$101,000). Additionally, the address’s BNB holdings were apparently liquidated six days ago, with an estimated loss of ~$73,000.

Beating AI News: Elorian AI, founded by former Google DeepMind researcher Andrew Dai, has launched its first visual reasoning model Glimpse, designed to handle actions, spatial relationships and object changes in videos. Glimpse can recheck frames during inference, zoom in on local areas, revisit key clips, and run code to verify details when needed. For example, to determine if a train is slowing down, it tracks the position of the same window across frames to compare displacement, rather than guessing based solely on video content. While similar frame cropping and zooming technologies have existed before, Elorian has made repeated checks of visual evidence a core capability of the model. Elorian’s internal tests show Glimpse outperforms GPT-6 Astra in four benchmarks: scientific chart reasoning, short video description, object counting and robot operation event recognition; it ties Astra in one test and lags in six others. Notably, Glimpse scores 56.68 on the comprehensive video understanding benchmark Video-MME v2, lower than Astra’s 64.06. These results are self-reported by the company and have not yet been independently verified. Glimpse is fine-tuned on an open-weight model. Elorian states its small team uses just hundreds of GPUs. The model’s API charges $0.60 per million input tokens and $3 per million output tokens, both 94% lower than Astra’s rates. Currently, the model is available as a research preview API, with its weights not yet made public. Elorian previously closed a $55 million seed round, valuing the company at $300 million; investors include Nvidia and former Google Chief Scientist Jeff Dean.

Trader 0xC7Fd opened a 2x long on 20.5M $AGT ($1.63M) on @Aster_DEX and is now sitting on over $1.26M in unrealized profit, with a 155% return.

CZ responded in a post to his recent deletion of a large number of old cryptocurrency-related retweets, stating he was merely testing an AI tool and cleaning up old reposts as a side task, adding "probably no one was looking anyway". CZ emphasized he did not delete any original content and urged the public not to overinterpret the incident. Earlier, community users spotted CZ deleting a large volume of old retweets, most of which were crypto-related, sparking discussions. The move appears to be a personal timeline cleanup rather than carrying any market signals.

Trader 0xC7Fd opened a 2x long on 20.5M $AGT ($1.63M) on @Aster_DEX and is now sitting on over $1.26M in unrealized profit, with a 155% return.

Beating AI News: Perplexity has open-sourced two multimodal decision models — pplx-decider-v1.2 (27B) and pplx-decider-flash-v1.2 (4B) — based on Qwen3.8-27B and Qwen3.5-4B respectively. Unlike models that generate long-form responses, these two models perform classification, scoring, and judgment based on text or images, suitable for use cases such as agent task routing and content moderation. Compared to the previous generation which required per-question inference, the new models allow multiple questions to share a single input, completing judgments in one inference step. Perplexity states that inference speed is up to 5 times faster in some scenarios. However, the 4B version may experience accuracy loss when combining multiple questions, so the official recommendation is to default to per-question processing. In the public test of Decision Index 0.3, the 27B model’s score rose from 62.30 to 68.09 compared to the prior generation, while the 4B version ranks first among models of the same scale. It should be noted that the new models incorporated official training datasets from some evaluation tasks during training. Both models are licensed under Apache 2.0, and integrated with Perplexity API and Vercel AI Gateway. API pricing is $0.02 per million input tokens for the 27B model and $0.01 per million input tokens for the 4B model, with free output tokens.