Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Each day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.
Fresh AI products Click to learn more:https://app.aibase.com/zh
1. OpenAI may launch a permanent AI assistant codenamed "O", expected to be officially released on September 29th
OpenAI plans to launch a new permanent AI assistant codenamed "O" at the OpenAI DevDay. This assistant may have its own digital identity and can run continuously. Many core details are still unclear, such as task scenarios, system permissions, and release time.
AiBase Summary:
🧠 OpenAI will launch a permanent AI assistant codenamed "O", capable of continuous operation.
🌐 This assistant may have an independent digital identity, breaking through traditional conversation modes.
📅 Specific features and official release date are yet to be disclosed by the official.
2. Claude Sonnet 5.5 configuration identifier appears and starts gray-scale testing, performance approaches GPT-6 Astra
The configuration identifier for Claude Sonnet 5.5 has appeared and started gray-scale testing. Its performance is impressive, not only surpassing GPT-6Sol but also approaching GPT-6Astra in some advanced coding and agent capabilities. Its pricing strategy is highly competitive, bringing cutting-edge capabilities to low prices, seen as a precise countermeasure against OpenAI DevDay.
AiBase Summary:
Claude Sonnet 5.5's configuration identifier "claude-sonnet-5-5" has appeared in the official configuration and started gray-scale testing.
Sonnet5.5 outperforms GPT-6Sol in performance and even approaches GPT-6Astra in some advanced coding and agent capabilities.
Its pricing strategy is highly effective, with input costs as low as $2 per million tokens and output at $10, with cache reading at just $0.20.
3. MiniMax's latest text model M3.1-Flash-Preview officially launched, with official distribution of quota reset cards and double bonus points
MiniMax's latest text model M3.1-Flash-Preview was officially launched, with the official distribution of quota reset cards and double bonus points to provide developers with more efficient and reliable development tools and more promotional activities.
AiBase Summary:
🧠 The M3.1-Flash-Preview model is fully launched, improving development efficiency.
💰 The official distributes quota reset cards and free token benefits, reducing usage costs.
📅 A limited-time sign-in double bonus points activity has been launched to enhance user engagement.
4. Zhipu ZCode launches a new round of compensation: gives paid users 8 reset cards, and has completely removed the code snapshot upload path
Zhipu ZCode launched a new compensation plan for previous code data upload controversy, while disclosing technical security improvement progress to rebuild developer trust.
AiBase Summary:
📌 ZCode launches a new compensation plan, giving paid users 4 weekly quota cards and 4 5-hour quota cards.
🔐 Zhipu has completely removed the code snapshot upload path and confirmed that all cloud data has been deleted.
📈 The competition in AI programming tools is shifting from model capabilities to data sovereignty and engineering security infrastructure.
5. The open-source large model landscape is reshaped! Xiaomi MiMo-V2.6-Pro makes a strong comeback to the top ten of Code Arena, performance rivals the first tier internationally
Xiaomi's new multimodal flagship large model MiMo-V2.6-Pro performed impressively in the Code Arena: WebDev test field, successfully returning to the top ten upon its debut and ranking among the top three in the open-source weight model under the MIT license.
AiBase Summary:
🧠 MiMo-V2.6-Pro performed well in the Code Arena: WebDev test field, successfully returning to the top ten upon its debut.
📈 The model scored 1628 points in AutoEval, matching Claude Fable5(High) and slightly surpassing Hy4-preview.
🚀 Compared to the previous generation MiMo-V2.5-Pro, it achieved a significant 153-point improvement in a single iteration.
6. The first end-side Agent flagship model touches the safety red line? Doubao AI denies involvement in game cheating
The article discusses the abnormal issues encountered when running "Honor of Kings" on the Nubia NaviX Ultra, pointing out that this may involve conflicts between AI end-side system permissions and game anti-cheating mechanisms. Doubao Mobile Assistant responded that it did not interfere with Tencent games and suggested users pause login or file appeals through official channels. The article also mentioned that the device is the world's first AI agent mass-produced flagship developed jointly by ZTE and ByteDance, with its technical features and market performance worth attention.
AiBase Summary:
🎮 Nubia NaviX Ultra displayed an "abnormal device environment" prompt while running "Honor of Kings," attracting attention.
🚫 Doubao Mobile Assistant denied any misconduct, emphasizing that it did not interfere with Tencent games and suggesting users pause login.
🚀 This device is the world's first AI agent mass-produced flagship developed jointly by ZTE and ByteDance, with notable technical features.
7. Google tests Gemini's direct purchase of Flipkart goods internally, fully entering the in-app transaction ecosystem
Google is accelerating the evolution of its AI services from simple "product discovery" to "transaction completion." Recent news shows that Google is conducting a new test in the Indian market, allowing users to directly purchase Flipkart goods through Gemini and AI Mode within the AI interface. The test is currently limited to certain users, and the product categories are mainly limited to a few areas such as smartphones, electronics, and phone accessories. Users can see the purchase button directly and click to seamlessly enter the checkout process without leaving the current AI interaction interface. To ensure large-scale business expansion, Google has set a clear timetable and plans to expand the test scope further in late October to precisely catch the important shopping season in the Indian market. Notably, Google not only launched a general business agreement but also made a significant investment of about $350 million to invest in Flipkart, laying a solid foundation for deep integration in local e-commerce. However, this direct purchasing function is still in its early stages, and not all retailers on the platform have joined the direct purchasing option, with the veteran e-commerce giant Amazon not yet included in this test system. As the internal closed-loop ecosystem gradually improves, the monetization capability of large models in the e-commerce sector will face new challenges.
AiBase Summary:
📱 Google tests the direct purchase of Flipkart goods within Gemini in India
🛒 Limited to certain users and specific product categories, supports seamless checkout process
📈 Google plans to expand the test scope to catch the shopping season
8. NVIDIA releases 100 million parameter free model Nemotron3, supporting real-time voice segmentation for 8 people
NVIDIA released a free AI model named Nemotron3Diarization with approximately 100 million parameters, focusing on voice segmentation clustering tasks. The model topped the VoiceArena voice segmentation benchmark test with an error rate of 14.72%, significantly outperforming the second-ranked system. The new model supports four levels of dynamic audio buffer settings ranging from 0.32 seconds to 30.4 seconds, accurately identifying speakers at any moment in a conversation, suitable for various application scenarios.
AiBase Summary:
🧠 The model has a large number of parameters and supports high-precision voice segmentation identification.
📊 Performed well in the VoiceArena benchmark test, with an error rate lower than industry average.
🔄 Supports dynamic audio buffer settings, balancing latency and accuracy.
