flash

Seedance announced the launch of the video generation model Seedance 2.5 flash, which supports generating videos up to 30 seconds long in a single instance, allowing for a maximum input of 10 images, 5 video segments, and 5 audio segments, providing resolutions of 480P and 720P.The official statement indicated that the generation speed of this model is twice that of Seedance 2.5; for example, the generation cost for video generated from text and images at 720P is approximately 0.99 yuan/second, which is about 34% lower than Seedance 2.5.Seedance 2.5 flash will gradually be available on platforms such as Jimeng, LibTV, Updream, Creao AI, and TapNow starting today, with the API expected to go live on November 15 on Volcano Ark.

According to IT Home, Baidu's Wenxin large model has released ERNIE 5.5 Flash Preview, which is the first post-training version based on the previously released Base model. This model has a total of 280 billion parameters, with 14 billion active parameters, and supports 1 million contexts. The official claims it performs excellently in programming, 3D modeling, knowledge work, content creation, and more.ERNIE 5.5 Flash Preview balances intelligence level and inference efficiency, achieving competitive scores in multiple benchmark tests while maintaining a fast decoding speed. The model API has been launched on the Qianfan platform, and the weights will be open-sourced before November 25.Baidu stated that this model is an intermediate version in the post-training process. Although it has achieved competitive performance among models of the same size, there may still be some capability deficiencies due to insufficient training. Post-training is still ongoing and will be optimized in the subsequent official version.

OKX will launch the Midnight (NIGHT) Flash Earn Lite event on November 11 at 15:00, where users can subscribe to BTC, SOL, OKB, or NIGHT to share a reward of 12,500,000 NIGHT.The event ends on November 24 at 15:00.

Base has disabled Flashblocks in the Denim upgrade, switching to generating a complete standard block every 200 milliseconds. Denim was activated on Base Sepolia on October 21, 2026, at 18:00 UTC, and on the Base mainnet on November 5 at 18:00 UTC. After activation, Base stopped producing Flashblocks on both networks, and Flashblocks subscriptions, WebSocket streams, and pending states will no longer return data.The official documentation states that endpoints like wss://mainnet.flashblocks.base.org/ws will still accept connections but will no longer push any data. Developers need to replace eth_subscribe("newFlashblocks") with eth_subscribe("newHeads"), replace pendingLogs with logs, and for newFlashblockTransactions, switch to pulling transactions by standard blocks after using newHeads; RPC calls using the "pending" block tag should be changed to read the latest standard block and state.Regarding node configuration, mainnet nodes must be upgraded to base/base v1.5.1 or higher, and Sepolia nodes must be v1.5.0 or higher, while removing settings such as RETH_FB_WEBSOCKET_URL, --flashblocks-url, --max-pending-blocks-depth, and --flashblocks.ping-interval. Additionally, validity transactions with the flashblock_index predicate will be rejected, and it is recommended to use block_number to limit the range.

A user stated that on November 8, 2026, while exchanging small cryptocurrencies for USDC on Binance, their account experienced 5 unauthorized flash exchanges within 97 seconds. The transactions they listed included 2.331 BTC exchanged for USDC, as well as two rounds of back-and-forth exchanges totaling approximately 260,000 USDC and EUR, which collectively reduced about 10,044 USDC, with each round close to 2%.A user claimed that Binance customer service insisted that the related transactions originated from their commonly used iPhone but did not provide sufficient records to prove that the user actually authorized them, nor did they adequately explain the exchange rate differences. They publicly raised three questions, including who initiated the transactions, why the price difference was significant, and who benefited economically, and published the complaint number CC10319344, requesting intervention from He Yi, CZ, and the Binance security team. They added that there were a total of 6 transactions in the screenshots, and only the EDEN exchange for USDC at 02:21:13 was operated by them, while the other 1 BTC exchange and 4 USDC and EUR exchanges were neither initiated nor authorized by them, with the EDEN exchange they operated occurring between the aforementioned 5 transactions.

According to China Blue News, on November 6, at the 2026 World Internet Conference in Wuzhen, Sohu founder, chairman of the board, and CEO Zhang Chaoyang admitted that he "missed the mobile internet." In the face of major companies heavily investing in large models, he explained that the outside world perceives Sohu's attitude towards AI as "calm," but "it is actually a conscious choice," because "large models require resource competition, which we cannot afford."When discussing algorithms and social interaction, Zhang Chaoyang stated that he does not oppose algorithms, but in product design, he chose not to pursue algorithmic recommendations, believing that some users would prefer to obtain information through real human social interactions. Regarding the short drama sector, he judged that the popularity of short dramas in China has not been replicated in the U.S. market, where American users still prefer long dramas and movies, saying, "The phenomenon of short dramas may be fleeting, or it may be a part of the future." He revealed that he does not watch short dramas, let alone AI short dramas.

Baidu has open-sourced the ERNIE-5.5-Flash-Base model weights. Following the release of the ERNIE 4.5 series in June 2025, Baidu has once again open-sourced a large model with hundreds of billions of parameters after approximately 16 months, during which the ERNIE 5.0 and 5.1 versions were not open-sourced. The ERNIE 5.5 Flash focuses on inference efficiency, with a total of 280 billion parameters, activating only 14 billion parameters for each generated token, and using 33.5 trillion tokens for zero pre-training.This model supports a maximum inference context of 1 million tokens. Baidu has redesigned the long text processing mechanism, with 37 out of 43 layers compressing every 128 historical tokens into a single cache record, while the remaining 6 layers retain fine retrieval capabilities for the original content, reducing the memory and computational overhead of long text processing. The official evaluation was conducted in conjunction with DeepSeek-V4-Flash-Base, Tencent Hy3-preview-Base, and Xiaomi MiMo-V2.5-Base, ranking first in 24 out of 30 benchmark tests.What is being opened this time is a pure text Base model that has not undergone intermediate training or post-training, mainly for developers to continue domain fine-tuning, instruction training, and reinforcement learning. Baidu stated that the post-trained ERNIE 5.5 Flash will be released subsequently. The license is stricter than the previous generation; the ERNIE 4.5 used the Apache 2.0 protocol, while 5.5 Flash has switched to a custom license that does not authorize use in the European Union, the United Kingdom, and South Korea; for related products or services that have generated over 50 million dollars in revenue in the past 12 consecutive months, separate written authorization from Baidu must be obtained before commercial use.

The liquidity service provider Quote Markets team member filo published a review of the XPL flash crash event on Hyperliquid. During the flash crash of XPL on Hyperliquid, 16 wallets associated with funds and trading activities withdrew approximately $2,038,600 in margin before the price plummeted, leading to HLP taking over a large number of XPL long positions.HLP ultimately took over about 160 million XPL and completed the sale within approximately 75 minutes, incurring a loss of about $555,787, equivalent to 55.6% of the clearing vault's own funds. Investigations showed that 14 of the 16 wallets had connections in terms of funding sources or trading activities, with 12 wallets withdrawing funds in advance, contributing to 90.7% of the positions taken over by HLP. However, the investigation did not find direct evidence that these wallets caused the XPL crash, nor could it confirm whether they hedged on other trading platforms.It is worth noting that these 16 wallets did not profit from this event on Hyperliquid, collectively incurring a loss of about $711,000. Additionally, XPL had already dropped about 6% before the first liquidation occurred on Hyperliquid, so the liquidation was not the starting point of this crash, and the initial active selling volume of Binance perpetual contracts was also significantly larger.

According to on-chain data monitoring, the flash crash of XPL price caused Hyperliquid's HLP treasury to incur a loss of approximately $556,000.Previously, 16 wallets funded through centralized exchanges had been continuously building long positions in XPL since October 29, holding a total of about 269 million XPL, with a nominal exposure of approximately $23 million. Among them, the funds of 14 wallets were transferred from Binance about two months ago, with 12 wallets withdrawing approximately 2.02 million USDC before the price drop to reduce margin exposure.As XPL rapidly fell from $0.086431 to $0.069429, it triggered a chain liquidation of long positions. After the liquidation, the HLP treasury passively took over approximately 160.39 million XPL long positions, with an average holding cost of $0.07592, and a nominal value of about $12.18 million. The HLP treasury ultimately closed at $0.0725, recording a loss of approximately $556,000.

According to official news from OKX, the new round of the OKX Flash Earn USDT event has launched, setting up two subscription pools for BTC and ETH, with a total prize pool of 1 million USDT. Among them, the BTC pool has 410,000 USDT, and the ETH pool has 590,000 USDT.Pre-subscription has been open since October 31 at 15:00, during which participants can earn simple interest on their tokens. Token rewards will only begin to be calculated after the official start of the event. The event runs from November 3, 2026, at 15:00 to November 8, 2026, at 15:00 (UTC+8).

According to a report by Blue Whale News, at the Tencent Global Digital Ecosystem Conference, Tencent Group Senior Executive Vice President and CEO of the Cloud and Smart Industry Group, Tang Daosheng, mentioned in his speech that they will release the new model Hy4 and Hy4 flash.Compared to the released Hy4 preview, Hy4 enhances the capabilities for various long-range and real productivity tasks, improves the efficiency of the thinking chain, and introduces visual understanding for the first time. Hy4 flash is positioned as a practical model that offers cost-effectiveness.

Binance will launch trading pairs for bStocks tokenized securities APPB, BSPB, GEVB, SKDDB, and SKUUB on October 28 at 20:00. Users can exchange bStocks for BTC, USDT, or other supported tokens without fees within one hour after the launch. In addition, the zero maker fee promotion for all bStocks trading pairs has been extended to December 1 at 07:59.

On October 26, B.AI announced a limited-time benefit, with the latest MiniMax model MiniMax-M3.1-Flash (Space Bunny) available at 50% OFF on the B.AI web platform, Android App, and API. This model is designed for complex agent tasks, software engineering, visual reasoning, and professional analysis, helping developers efficiently complete complex tasks from code development to multi-step automated workflows.The event runs from October 26, 17:00 to November 2, 16:59 (SGT), allowing users to experience the popular new model at a lower cost and flexibly manage AI invocation costs. B.AI stated that it will continue to integrate high-quality models from around the world, providing a richer selection of models and more competitive prices through a unified entry point. Users are welcome to log in to chat.b.ai/chat to experience it immediately.

OKX will launch the CT (Concrete) Flash Profit Lite event from October 28 at 15:00 to November 3 at 15:00. Users can share a reward of 1,000,000 CT by subscribing to BTC, ETH, OKB, or CT.During the event, users must complete identity verification to participate. Sub-accounts and institutional users are not allowed to participate.

MiniMax announced the open-source release of the MiniMax M3.1 Flash model weights on October 23, with a total of 428 billion parameters and 23 billion active parameters. Compared to the previous generation M3 model, the M3.1 Flash has achieved a comprehensive improvement in model capabilities. According to official data, within 72 hours of the model going live, the number of tokens processed by real users exceeded one trillion.On the technical side, M3.1 Flash improves inference efficiency through optimizations such as full model sparse attention, quantization-aware training (QAT), and power-aware scheduling. In terms of ecosystem, MiniMax collaborates with chip manufacturers such as Huawei Ascend, T-head, Kunlun Core, Moore Threads, Biren, Haiguang, Muxi, Tiandao, NVIDIA, AMD, and cloud platforms like Together AI, Fireworks, Nebius, Baseten, Modal, DigitalOcean to promote access on the first day of open source. vLLM, SGLang, Zhiyuan FlagOS, as well as Hugging Face, Modao, and other communities simultaneously support the deployment of the model across different chips and development environments.

The AI model evaluation platform Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has ranked first among open-source models in Agent Arena and 11th overall. Based on approximately 7,500 real usage sessions, this model shows a comprehensive net improvement of about 5.1% compared to the average model, with a median single-task cost of about $0.20.This version ranks first among open-source models in four metrics: task success confirmation, user positive and negative feedback, reliability of Bash command execution, and guideability, with the task success confirmation metric ranking 8th overall.

On October 19, the B.AI platform announced that MiniMax's MiniMax M3.1 Flash (also known as Space Bunny) is officially online, with API and Web Chat both available for a synchronized experience. This model is designed for complex office automation, software engineering, visual reasoning, and professional analysis, supporting multi-step task execution, front-end implementation, and presentation creation, helping developers and professional users efficiently transform ideas into tangible results. The model supports a 1M Token context window and up to 512K output Tokens, with standard prices as low as $0.15 per million input Tokens and $0.60 per million output Tokens, balancing complex task processing with high cost-effectiveness.

MiniMax announced that M3.1 Flash has officially been released, and the earlier version, Space Bunny, which topped global rankings and processed over 200T tokens, is its predecessor.M3.1 Flash achieved scores of 62% on Automation-Bench, 30% on ALE, 74% on DeepSWE, and 91% on BabyVision in tasks such as Cowork, Coding, and Visual Reasoning.

On October 19, the B.AI platform announced that DeepSeek V4.1 Flash has seamlessly upgraded to the latest V4.1 Flash-1017 version, further optimizing response speed, inference quality, and operational stability, bringing a more efficient and intelligent AI experience to developers and users.This upgrade continues the seamless switching method, allowing users to keep the same model ID, and existing API calls and workflow configurations can remain unchanged, enabling direct access to the new model capabilities, significantly reducing migration and adaptation costs. B.AI stated that it will continue to follow the iteration of cutting-edge models, providing developers with more stable and cost-effective model services.

The trading execution platform Tread.fi CEO DYJ published an analysis of the flash crash of AI concept stocks' perpetual contracts in the US stock market early in the morning, stating that the sell-off began at Binance. A trader liquidated approximately 2 million dollars of SanDisk long positions, triggering a sharp decline in other markets due to the thin depth of the order book.According to the transaction timeline listed, a large amount of active selling first appeared on SNDK at Binance, with a transaction amount of about 407,000 dollars in the first second, followed by about 1,030,000 dollars in the next second, and MU also began to see selling in the same second. Hyperliquid then saw sweeping orders, with about 10 addresses submitting IOC sell orders concentrated on SNDK, SOXL, DRAM, SKHX, SKHY, and MU within 1.8 seconds, followed by a second round of synchronized selling. These orders were all used to open new short positions, rather than just reducing positions.In the minute of the first concentrated selling, the transaction amounts for SNDK and MU on Binance were approximately 21.9 million dollars and 5.1 million dollars, respectively, while the corresponding transaction amounts on Trade.xyz were about 3.1 million dollars and 300,000 dollars. The transaction amount for MU on Hyperliquid subsequently expanded to 10.8 million dollars and 9.5 million dollars within the next two minutes, which DYJ attributed to chain liquidation.