Lookonchain APP

App Store

Tencent's Hunyuan Hy4 overhauls its underlying layer, integrating DeepSeek's sparse attention and Zhipu's IndexCache.

2026.08.28 15:40:55

Beating AI Express News: Tencent’s Hunyuan Hy4 preview not only scales up the model size but also overhauls its underlying architecture. The previous Hy3 generation used traditional full attention, while Hy4 adopts Gated DSA, the gated version of DeepSeek Sparse Attention. When processing extremely long texts, the model no longer recalculates all previous content; instead, it first selects the most relevant parts for focused processing. Hy4 also adds IndexCache to DSA. Originally, each layer of the model had to re-judge which content is most important, but now some layers can directly reuse earlier selected results, cutting down on significant redundant computation. Tencent explicitly stated that this design is inspired by DeepSeek and GLM. Another change is the residual structure: traditional Transformers can be roughly viewed as having only one information backbone, while Hy4 uses iHC to expand this to four backbones, enabling the model to retain more information simultaneously in deep networks. The number of MoE (Mixture of Experts) experts has also increased from 192 in Hy3 to 256 routing experts plus 1 shared expert. Each token selects 8 routing experts and always passes through the shared expert.

Relevant content

A trader engaged in $2.84 billion worth of wash trading on Papertrade, suffering a net loss of $637,000.

According to Lookonchain's monitoring, a trader has accumulated a $637,000 loss on Papertrade, with a cumulative notional trading volume of $2.84 billion. The trader incurred a $1.71 million trading loss and received 136 million PAPER tokens. Additionally, the trader earned $1.07 million in USDC rewards via staking, of which $943,000 has been claimed while $130,000 remains unclaimed. Overall, the trader’s net loss stands at $637,000.

1 seconds ago

Microsoft’s Stunning Reversal: Stock Rebounds Over 50% From Its June Lows, Eyeing All-Time High

As Microsoft (MSFT.O)’s four-month stock rally continues to accelerate, the company’s shares are on track to hit their first closing all-time high in nearly a year. Amid growing market confidence in Microsoft’s competitive position in artificial intelligence (AI), the stock has risen over 50% since its June low, adding roughly $1.5 trillion to its market capitalization. This milestone marks a dramatic reversal for Microsoft’s share price. Just months ago, the tech giant was among the worst-performing stocks in the S&P 500, as investors worried that its massive AI infrastructure spending was failing to deliver sufficient returns and that the software provider could face disruptive shocks. The stock plummeted 17% in June, its worst monthly performance since 2000, erasing nearly $600 billion in market value. The turning point came on July 29, when Microsoft released earnings showing its cloud computing business grew at its fastest pace in four years. The next day, the stock surged 16%, kicking off a strong rebound that led to Microsoft’s best quarterly performance since 1998. Following the July earnings release, at least two institutions upgraded Microsoft’s stock rating to "Buy", citing confidence in the company’s ability to capitalize on growing demand for AI services.

1 seconds ago

Trader 0x26b6 loses $637K on Papertrade despite $1.07M in USDC staking rewards

Trader 0x26b6 lost $637K on @papertrade_xyz. He lost $1.71M from trading and earned 136M $PAPER tokens. He earned $1.07M in USDC rewards through staking, with $943K already claimed and $130K still unclaimed. His net loss is $637K.

1 seconds ago

W dropped over 30% in 24 hours, now trading at $0.0181.

According to HTX market data, W is currently trading at $0.0181, down 30.45% over the past 24 hours.

1 seconds ago

Interactive Brokers launches Ethereum staking service, powered by ZeroHash.

According to market sources, Interactive Brokers has launched an Ethereum staking service powered by ZeroHash.

1 seconds ago

Renowned trader Killa: Bitcoin needs to hold the $80,000 level and form a higher low; otherwise, the correction could deepen.

Well-known crypto trader Killa posted that for Bitcoin to evolve along its current price structure, it needs to form a higher low above $75,500, ideally holding the low near $80,000. This mirrors Bitcoin’s 2023 price action: back then, it posted a higher high on the weekly timeframe, then retested and defended its original range. This may explain why recent lower-timeframe moves have been so prone to triggering stop-losses, aggressive, and marked by algorithmic trading activity. Bitcoin once recovered its internal structure in the $82,000 range, then retested, broke below the structure again, and later reclaimed it. The entire market seems to be deliberately luring traders into positions. Admittedly, the initial dip to the $81,000 range was structurally unfavorable, but a large number of short positions were concentrated at lower levels at that time, which may account for the subsequent price reaction. Another notable detail: during Bitcoin’s decline from $84,100 to $81,000, only around $300 million in long positions were liquidated; whereas in the prior drop from $86,800 to $80,600, long liquidations exceeded $2 billion. The sharp fall in long liquidation volume indicates the market’s reliance on long positions may be lower than before. Current price action is quite delicate. Bitcoin needs to form a higher low at its current level to kick off a new rally; otherwise, this correction could push prices below $80,000. For now, the so-called "crash protection team" has done a solid job of propping up prices, but it’s also constantly disrupting lower-timeframe internal consolidation ranges.

1 seconds ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano