Lookonchain APP

App Store

Qwen 3.8-Max Gets Further Upgrade: All 8 Programming Test Benchmarks Improve, Some Outperform Opus 5

52 minutes ago

Beating AI Express: Alibaba’s Qwen has upgraded its flagship model Qwen3.8-Max, which was launched just one month ago, to Qwen3.8-Max-0902. The parameter scale remains unchanged at 2.4 trillion parameters with a 1 million-token context window. This upgrade focuses on post-training for coding and collaboration, with a priority on improving programming, complex workflows, and long-horizon agent tasks. In Alibaba’s published comparison table, the 0902 version outperforms the original in all 8 programming benchmarks. Specific improvements include: TerminalBench 3.0 rising from 11.3 to 29.0, DeepSWE 1.1 from 56.6 to 69.3, QwenSWEbench V2 from 55.1 to 70.0, and professional task benchmark JobBench from 53.4 to 64.0. Compared with Claude Opus 5, Qwen3.8-Max-0902 still lags in most programming projects, but surpasses the competitor in three benchmarks: MLS-Bench-Lite, SWE-Atlas QnA, and QwenSWEbench V2. The four multimodal benchmarks only saw gains of 0.4 to 3 points over the original, making it clear that this upgrade’s core focus is enhancing the model’s coding and task execution capabilities for agents. All the above results are currently Alibaba’s internal tests. Qwen3.8-Max-0902 is now available on the QwenCloud API, with pricing unchanged: $2 per 1 million input tokens and $6 per 1 million output tokens.

Relevant content

UNI rose around 14% in the past 24 hours, up more than 150% from its bear market bottom.

According to HTX market data, UNI has gained around 14% in the past 24 hours, currently trading at $6, and has rallied more than 150% from its bear market bottom. Per DeFillama data, Robinhood Chain’s total DEX trading volume over the past 24 hours stood at approximately $1.673 billion, of which Uniswap-related trading volume hit $1.232 billion, accounting for roughly 73.6% of the chain’s total DEX volume. In terms of fees, Uniswap-related protocols generated about $9.24 million in fees on Robinhood Chain over the past 24 hours: V4 contributed around $6.83 million, while V3 added roughly $2.24 million. Under DeFillama’s metrics, Uniswap-related protocols earned approximately $416,000 in revenue on Robinhood Chain over the past day, primarily from Uniswap V3.

8 minutes ago

Elon Musk: Grok 4.7 to Be Released in Ten Days

Beating AI Express News: Elon Musk has just confirmed that Grok 4.7 will be released in 10 days. As he previously disclosed, the new model’s parameter count will rise from 1.5 trillion parameters in Grok 4.6 to 2.1 trillion parameters. Initial training has been completed, and the model is currently undergoing supplementary training using a large amount of SpaceX data.

8 minutes ago

A crypto whale sent 95 Bitcoin to Galaxy Digital, likely for over-the-counter (OTC) sale.

According to monitoring by Onchain Lens, a whale has sent 95 Bitcoin (valued at approximately $7.32 million) to Galaxy Digital, likely for an over-the-counter (OTC) sale. https://www.panewslab.com/zh/articles/01a06031-04f9-7595-80b4-a81e5827df31

8 minutes ago

Altcoins post minor broad gains, with CRV up 9.36% in the past 24 hours.

According to HTX market data, as Bitcoin and Ethereum pull back amid geopolitical turmoil, some altcoins have seen a minor broad rally. Among them: FIL is up 14.6% in 24 hours, currently trading at $0.79; UNI is up 10.9% in 24 hours, currently at $6; CRV is up 9.36% in 24 hours, currently at $0.39; PYTH is up 7.33% in 24 hours, currently at $0.054; AR is up 7.17% in 24 hours, currently at $2.25; SKY is up 6.9% in 24 hours, currently at $0.072; ENA is up 6.52% in 24 hours, currently at $0.16.

8 minutes ago

A self-proclaimed trader operating under the crypto handle "喜欢梭哈allin.eth" was liquidated after going long on gold, with total losses reaching $1.27 million.

According to monitoring by TradingBeats (formerly Hyperinsight), a whale self-named "喜欢梭哈 allin.eth" was liquidated on its long gold position today. The address opened a long GOLD position on August 31 at an average price of $4,422.3, using 20x isolated leverage, with a position size of nearly $2 million. During today’s gold price drop, it first voluntarily reduced its position by 154.20 units near $4,311, incurring a loss of approximately $17,200. The remaining 298.05 units were then liquidated by the system at around $4,290.06, with a liquidation value of about $1.279 million, leading to an additional loss of roughly $39,400. The entire gold trade resulted in a total loss of approximately $56,600. As of press time, the address’s gold position has been reduced to zero, but the account is not fully liquidated: it still holds a long xyz:CBRS position worth around $187,000; its cumulative historical PnL for perpetual contracts remains approximately positive $308,000. Additionally, the on-chain wallet controlled by this address corresponds to the public Polymarket account "Azen". It once invested roughly $999 in a very early bet that "the Federal Reserve would cut rates by more than 50 basis points after its September meeting", ultimately netting a profit of about $1,890. However, this is currently the only completed public transaction by Azen on Polymarket. Address: 0xfe40fa3decfc6dc555f4ad7736d0dd65ccbd4743. On-chain perpetual and address analysis tool TradingBeats is now live, supporting real-time viewing of Hyperliquid data, enabling full deep analysis from address tracing to whale operations.

8 minutes ago

Anthropic locks Claude’s chain of thought: it is invalidated if even a single line of context is modified, specifically to block model distillation.

Beating AI News Flash: Anthropic adds "context locks" to Claude’s hidden reasoning. Fable 5.1’s encrypted reasoning generated via API must now be returned exactly as it was when created, alongside the system prompt, tools, and historical messages used to generate it. If any preceding content is altered, the API will either throw an error or discard the reasoning segment entirely. This change is designed to prevent model distillation. Earlier, researchers found that while Claude’s encrypted reasoning is unreadable to users, it can be fed back to Anthropic’s compatible models. Attackers could prompt Opus to produce high-quality reasoning, then insert the encrypted block into the less secure Haiku model, tricking it into outputting Opus’s full reasoning process. This amounts to not only copying the large model’s final answer, but also stealing its step-by-step reasoning from the "scratch pad". When Anthropic launched Fable 5 in June, it already rolled out a fix by binding encrypted reasoning to the model, blocking it from being sent to smaller models like Haiku for decryption. Fable 5.1 now adds "conversation binding": as long as the model remains unchanged, modifying the preceding prompt, tools, or chat history will invalidate old reasoning. This lock is not yet fully enforced. It will only be mandatory for new API accounts created after August 31 when calling Fable 5.1; existing accounts are temporarily unaffected. Claude.ai, Claude Code, Cowork, and other Claude models are excluded from this update. Anthropic states that future new models will enable these rules by default for all users.

8 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano