parameter

Tencent Hunyuan has released and open-sourced the next-generation large language model Hy4 preview. This model has a total of 770B parameters and 49B active parameters, with a context length exceeding 1M. It demonstrates strong capabilities in real productivity tasks such as coding, office work, and science, firmly placing it in the top tier of open-source models. Hy4 preview significantly expands in model size, context length, and data scale, and enhances real-world performance through high-quality data co-built with Tencent experts in software engineering, gaming, finance, security, and deep collaboration with products like WorkBuddy.In software engineering, the model enhances understanding, planning, debugging, and verification capabilities for long-range development tasks, enabling the construction of complex front-end projects like a Three.js miniature town from scratch. In game development, it supports generating playable prototypes from a single sentence and can complete a full demo in Unity. In smart office applications, it can handle complex financial audits, filtering, analyzing, and delivering from multiple documents. In scientific research, it achieves acceleration in tasks such as molecular dynamics simulations and has initially formed a recursive self-improvement feedback loop.Hy4 preview can be experienced in Tencent products such as WorkBuddy/CodeBuddy domestic and international versions, Yuanbao, ima, and can also be accessed via API calls through Tencent Cloud Tokenhub and OpenRouter. WorkBuddy/CodeBuddy will launch a limited-time free activity for two weeks. Since the reconstruction of the infrastructure, the Hunyuan large model has iterated a major version approximately every two months, continuously optimizing through a preview-first and formal version-following approach.

Yesterday, Ledger disclosed details of the LSB 023 security vulnerability on its official website. Some applications built on the Ledger Secure SDK may still receive new APDU commands during the user screen confirmation, leading to inconsistencies between the parameters displayed on the screen and the final signature parameters.In cases where the attacker controls the APDU communication between the device and the host, after the user confirms the operation shown on the screen, the device may generate a signature with different parameters. Ledger stated that it has fixed the issue through application-level checks and the SDK layer, and released Ledger Secure SDK v26.6.1 on August 21. The relevant applications have been rebuilt and released. Users need to update the applications through Ledger Live; merely updating the device firmware is insufficient to complete the fix. However, Ledger claims there is currently no evidence that the vulnerability has been exploited.

Google has confirmed that it is gradually promoting the "goto" redirect parameter in search result links. When users click on a link, they will first go through a redirect address with the google.com/goto parameter, rather than directly pointing to the target webpage. This mechanism increases the difficulty for automated tools to batch parse and scrape search results through server-side redirection, in response to scraping behaviors from third-party tools and AI companies.SEO monitoring platform Nozzle's data shows that this change has nearly 100% coverage in multiple residential IP service provider environments.

According to Jinshi reports, the dark side of the moon Kimi announced that the company's first-generation trillion-parameter multimodal model Kimi K2.5 will end its service at the end of this month.

According to LatePost, ByteDance is discussing a large model with training parameters exceeding 50 trillion, surpassing Alibaba's Qwen 3.8-Max (24 trillion) and Moonlight K3 (28 trillion), making it the largest known plan in the country so far. This plan is still in its early stages and does not guarantee a final release. The new model is intended to be led by Xiang Liang, head of Seed Foundation, in collaboration with Shen Ke, who is responsible for the pre-training data of large language models. Seed is reorganizing, dividing responsibilities, and allocating resources based on this.Two weeks ago, ByteDance founder Zhang Yiming held a company-wide meeting with Seed head Wu Yonghui. Zhang reassured the team that training large models is inherently difficult and that it is acceptable to lag behind for a period of time, hoping to aim for the upper limits of intelligence and join the world's top tier. He acknowledged that programming is a key direction at present, advocating for the integration of Volcano Engine, Feishu, and Doubao resources to build computational power and data advantages, while reminding not to be led by a single hot topic. He praised Seedance's differentiated leadership and clearly opposed distillation, believing it is difficult to truly surpass and that AGI barriers should be built from a more fundamental level, stating that the company will continue to increase investment in AI.In the past six months, Seed's multimodal performance has been outstanding, with Seedance 2.0, Seedream, and others driving Volcano Engine MaaS, but the market response to the language model Seed 2.0 has been limited, and its lagging coding capabilities have affected the revenue structure. ByteDance has hired Guo Daye at a high salary to specialize in coding and has consolidated related resources. In the face of the industry's general trend of increasing model sizes, ByteDance hopes to achieve a leapfrog advantage with a larger scale while promoting the elimination of horse racing and breaking down departmental walls to concentrate efforts on tackling challenges.

According to the Financial Times, as Chinese companies continue to narrow the gap with the top AI labs in the United States, ByteDance is training an AI model that may be close in scale to Anthropic's advanced Mythos system. According to three insiders, this Chinese tech giant is currently in the early stages of training a large-scale model, which may have a parameter count reaching the trillions, approximately three times that of the largest model released in China, the Kimi K3 model under the Moonlight brand.One insider stated that ByteDance's model is currently undergoing pre-training, a phase that typically takes three to six months. If all goes well, it will then undergo fine-tuning and be released. The specific scale of the model will be determined later. Anthropic has not publicly disclosed the parameter scale of its models, but industry estimates suggest that its advanced Mythos 5 model has a parameter count of about 8 trillion, while Fable 5 has around 5 trillion.It is important to note that the number of parameters determines the model's basic capacity for storing information or memory limit, but the actual capabilities of the model also depend on other factors, including data quality and training methods.

According to a report by LatePost, ByteDance is discussing training a model with a parameter scale exceeding 50 trillion, surpassing Alibaba's Qwen3.8-Max (24 trillion parameters) and the K3 from Dark Moon (28 trillion parameters), making it currently the largest known model in terms of parameter scale in the country. Generally speaking, the larger the model scale, the higher the level of intelligence.However, this plan is still in the early stages and does not guarantee that the model will ultimately be released. The new model will be led by Xiang Liang, head of SeedFoundation, in collaboration with Shen Ke, who is responsible for the pre-training data of large language models. To this end, Seed is reorganizing, defining responsibilities, and allocating resources.

The Web3 fitness protocol BRT.Fit has released a transparency statement, currently in the product testing and ecosystem validation phase. Application functions and validation methods may be adjusted based on testing results. The test token tBRT is only used for economic model validation and does not represent the official token value, nor does it constitute an investment product or guarantee of future returns. Its "mandatory exercise protocol" is an incentive and commitment mechanism that users voluntarily join, helping to develop exercise habits through verifiable data, without physical compulsion.After the official mainnet launch, core functions such as token staking, rewards, and burning will be executed through on-chain smart contracts, with contract addresses and audit reports to be disclosed gradually. The project plans to support mainstream wearable devices such as Apple Watch, WHOOP, and Xiaomi, Huawei, and will continue to optimize the anti-cheating system. The project team reminds that digital assets carry significant volatility risks, participation does not guarantee returns, and users should refer to official announcements, remaining vigilant against counterfeiting and phishing activities.

According to Xinhua News Agency, as the 2026 World Artificial Intelligence Conference and the High-Level Meeting on Global Governance of Artificial Intelligence are about to be held, the new generation model Kimi K3 was released on the 16th, with a parameter scale of 28 trillion. This is currently the largest open-source model in the world by parameters, marking a new step in the development of artificial intelligence models in our country.Beijing Moon's Dark Side Technology Co., Ltd., which released the model, introduced that Kimi K3 natively supports visual understanding, has a context window of 1 million tokens, and is optimized for complex task scenarios such as software engineering, knowledge work, deep research, and multimodal understanding, further enhancing the processing capability of large models for complex tasks. It is understood that in evaluations, Kimi K3's comprehensive intelligence level is close to that of the world's leading closed-source models. A relevant person in charge of Moon's Dark Side stated that in the process of training the new generation foundational model Kimi K3, they used a self-developed underlying model architecture and accumulated a complete set of scientific model training methods.

According to the Financial Times, informed sources reveal that the Chinese AI unicorn company Moonshot AI plans to release a new large language model, Kimi K3, in the near future. This model has between 20 trillion to 30 trillion parameters, making it the largest AI model in China by parameter scale, and its performance is expected to surpass the flagship model Claude Opus 4.8 from Anthropic in mainstream benchmark tests (industry speculation suggests its parameter count is around 15 trillion to 20 trillion).Unlike the currently mainstream closed-source and expensive cutting-edge large models in the United States, Kimi K3 will be available as an open-weight model for users to download and modify for free, which may create competitive pressure for leading American labs like OpenAI and Anthropic. Currently, due to the rising service fees for large models in the U.S. (for example, Anthropic has announced a 50% price increase for Opus 4.8 in September), some overseas companies have begun to shift towards using more cost-effective Chinese open-source models.In terms of the capital market, informed sources indicate that Moonshot AI is preparing for a new round of financing, with the latest valuation expected to reach approximately $31.5 billion. Meanwhile, the valuations of other AI giants in China and the U.S. are also rising; DeepSeek is starting a new round of financing with an estimated valuation of about $71 billion, while Anthropic and OpenAI have reached valuations of $965 billion and $852 billion, respectively, in their latest round of financing. In response to the aforementioned release and financing rumors, Moonshot AI has currently declined to comment.

According to a report by ZeroHedge, JPMorgan predicts in a research report that Zhipu will release the next-generation flagship model GLM-5.5 in August, with the parameter count possibly exceeding one trillion (>1T).Zhipu just open-sourced the GLM-5.2 model with a total parameter count of 744B in mid-June, which has performed impressively in several long-range programming benchmarks. The upcoming trillion-parameter model release will become the next key indicator to assess whether Zhipu can continue to maintain an upward trajectory in its technology curve.

According to the official announcement, Bitget has adjusted the leverage, position tiers, and maintenance margin rate for the SPCXUSDT U-based perpetual contract trading pair. Based on the position limit, the leverage has been increased to a maximum of 100 times.Existing positions before the user update will be affected, and users can adjust the relevant parameters according to their asset allocation needs.

According to L2BEAT analysis, users of the Ethereum Layer 2 network Scroll were overcharged more than $50,000 in transaction fees within about four days last week. The reason is that the Scroll team manually adjusted two rate multipliers in their Gas price oracle six times within six days, increasing each by 2 to 10 times, cumulatively raising the L1 data cost parameter to 1,280 times the original benchmark.Approximately 139,000 affected transactions originally had a total cost of only about $280, but were actually charged over $50,000, with automated bots accounting for the vast majority. L2BEAT pointed out that the overcharging was not an issue with the sequencer, but rather stemmed from the team's multiplier adjustments executed through a multi-signature wallet. This incident has raised questions about whether Scroll had previously subsidized users with rates below cost to maintain activity. According to DeFiLlama data, Scroll currently has a TVL of only $24 million, down 96% from its peak of $585 million in October 2024. As of the time of publication, Scroll has not publicly responded.

ChainCatcher message, the Ethena Foundation confirmed on social media that the fee switch parameters set by the Risk Committee have met the requirements. It is expected that the fee switch for ENA will be activated after the Risk Committee approves the implementation details and completes the subsequent governance process.The Ethena Risk Committee is currently discussing the implementation details, which will be made public once finalized ------ at that time, ENA holders will be required to vote to confirm the proposed framework.

ChainCatcher news, according to the official announcement, JustLend DAO lowered the sTRX energy leasing APY on July 13, while maintaining the energy return fee mechanism. This adjustment is expected to significantly reduce users' leasing costs while enhancing the staking rewards for TRX holders.As a leading lending protocol in the TRON ecosystem, JustLend DAO is committed to optimizing the DeFi product experience and advancing the improvement of the financial infrastructure of the TRON network. The platform will continue to optimize parameter configurations to ensure the robust operation of the protocol and the safety of user assets.

ChainCatcher message, the Bitcoin staking protocol Babylon announced on platform X that the governance proposal to modify the Babylon Genesis chain parameters has officially launched. This proposal aims to adjust the unbinding fee for the second phase of staking from 100 sats/vbyte to 30 sats/vbyte. Voting is now open, and the deadline is 7 AM UTC on Monday, April 21.

ChainCatcher news, according to Jinshi reports, there are market rumors that the EU trade chief discussed parameters for reducing tariffs with the United States.

ChainCatcher message, according to a warning from the Slow Mist security team, "We have detected two suspicious transactions using the same method on the ETH and BSC chains. The attacked contracts may have failed to check the incoming parameters in their swap-like functions, allowing attackers to construct malicious tokens and profit. However, these contracts are not open source."Slow Mist reminds users to stay vigilant and pay attention to asset security.

ChainCatcher message, friend.tech posted on X stating that the Admin and ownership parameters have been set to 0x000...000 to prevent any future changes to its fees or functionality. This change will not affect the separate web client running on friend.tech, which will continue to operate normally. Currently, no fees from the smart contract or friend.tech will flow to the friend.tech development team's multisig.