Arthur Hayes: GPIF May Re-Examine Internal and External Asset Allocation; 'Money Printing Feast' Ahead

Hyperliquid Research Collective (HRC) released a report noting that after Hyperliquid opened third-party permissionless deployment for its HIP-4 prediction market layer on August 29, the platform’s trading volume grew rapidly. The average daily trading volume of HIP-4 in the first 28 days of August was around $545,000; following the deployment opening, single-day volume hit $1.97 million on August 31, with a 24-hour trading volume reaching $2.75 million, and the number of active traders rose from 1,256 to 1,841. The report points out that prediction market project Outcome has been the main beneficiary, currently accounting for nearly 85% of HIP-4’s total trading volume, and its $1 million trading incentive program further boosted liquidity growth. HRC attributes Hyperliquid’s core advantage to its unified account system: prediction markets can share the same account environment as perpetual contracts and HIP-3 assets, allowing users to hedge perpetual positions via prediction market contracts—an experience not currently offered by platforms like Kalshi and Polymarket. Sports prediction markets may become HIP-4’s largest growth area. During the recent World Cup, HIP-4-related markets recorded a cumulative trading volume of $189.5 million, accounting for around 3% of the global World Cup prediction market trading volume. However, HRC states that HIP-4’s current main limitation is not on-chain deployment, but regulatory access. The U.S. market involves regulatory frameworks from the CFTC, SEC, and other bodies, with sports prediction markets in particular likely triggering gambling-related regulatory scrutiny. HIP-4 has proven that permissionless deployment can rapidly expand trading scale, but whether it can further grow its market share will depend on the regulatory environment, recovery of the sports market, and future governance votes.

Beating AI Express: Right after Gemini 3.8 Flash launched, Google scored 73.7% on DeepSWE v1.1. The DeepSWE official leaderboard now lists the model at 74%±1%, placing it ahead of Claude Opus 5 and GPT-5.6 Sol. Just hours later, Meta unveiled Muse Spark 1.3, which notched a 75.4% score on the same DeepSWE benchmark, pushing the public benchmark score up by an additional 1.7 percentage points. The test tasks agents to independently complete long-cycle software engineering tasks from 113 real codebases. Both Google and Meta used the mini-swe-agent framework: Google ran the high inference profile, while Meta used the max profile. Meta's previous-generation Spark 1.2 only scored 55.0%, marking a 20.4 percentage point jump this time around. However, Meta's 75.4% score has not yet been added to the official DeepSWE leaderboard. On Artificial Analysis's Coding Agent Index, the limited-preview Spark 1.3 max scored 68 points, second only to Claude Opus 5; the currently publicly available xhigh version of the model scored 64 points.

According to monitoring by ai_9684xtpa, a trader who has previously executed multiple ETH swing trades bought 1,650 ETH at approximately $2,417 per unit 1 hour ago, totaling around $4 million. Statistics show that this address has completed 4 ETH swing trades since July, generating a cumulative profit of roughly $1.05 million.

Per TradingBeats’ monitoring, as of press time, an address starting with 0xc74 has been steadily accumulating HYPE. Today alone, the address has bought 90,772.69 HYPE tokens, totaling roughly $7.43 million in transaction volume. The address’s current HYPE spot holdings breakdown is: HyperCore spot account: 50,331.0493 tokens; HyperEVM native HYPE: 47,950.5328 tokens; Total: 98,281.5822 tokens, valued at approximately $8.0469 million. Its total remaining purchasing power stands at around 9,267.72 tokens. If the address deploys all its remaining USDC, its theoretical HYPE spot holdings could rise to about 107,549 tokens.

Beating AI News: As its AI cloud business expands, Bitcoin mining firm IREN is projected to generate around $17 billion in annual AI cloud revenue by 2030, with its cloud computing capacity set to expand to approximately 2GW. Bernstein forecasts that IREN will phase out its Bitcoin mining operations by the end of 2026. Meanwhile, the price of IREN’s three-year AI cloud contracts has risen by roughly 125% since last November, reflecting sustained growth in AI computing power demand and related pricing power.

Per GMGN market data, Robinhood Chain meme project NUDES currently has a market capitalization of roughly $14.27 million, with its token trading at $0.0142, a 257.5% gain in the past 24 hours, and a 38.72% rise over the last six hours. The token’s market cap once touched a historical peak of about $15.94 million, and has now retreated to around $14.27 million.