AI 新闻 —— July 15, 2026
PulseAugur 当天浮现的 20 条头条故事 —— 综合实验室、论文及开发者社区的信号进行排序。
-
OpenAI's GPT-5.6 Sol Pro solves 30-year-old statistics conjecture in 90 minutes
A University of Pennsylvania statistics professor utilized OpenAI's GPT-5.6 Sol Pro to solve a 30-year-old statistical conjecture related to the Benjamini-Hochberg method. The AI model achieved this feat in approximately 90 minutes, a task that proved insurmountable for its pred…
-
Xmax AI unveils X2.0 with real-time video interaction and replacement features
Xmax AI has launched its new real-time interactive AI video model, Xmax X2.0, which features upgraded real-time rendering capabilities. The model introduces functionalities such as CharX for real-time person replacement, ClothX for real-time clothing changes, and VibeX for real-…
-
Xmax AI unveils Xmax X2.0 with real-time video interaction
Xmax AI has launched its new real-time interactive AI video model, Xmax X2.0, which features upgraded real-time rendering capabilities. This new model supports functions like CharX for real-time person replacement, ClothX for real-time clothing changes, and VibeX for real-time s…
-
China's central bank to increase reverse repo operations; 7 new generative AI services approved
China's central bank is exploring ways to increase the frequency of overnight reverse repurchase operations to manage liquidity, according to a statement from Deputy Governor Zou Lan. This move is part of a broader effort to refine the monetary policy framework and guide market …
-
Meta unveils Muse Image, emphasizing safety in generative AI
Meta has introduced Muse Image, a new generative AI model designed for image creation, as part of its broader Muse ecosystem. This model is built with a strong emphasis on safety, incorporating systematic risk evaluations and structured red-teaming inherited from the Muse Spark …
-
Tencent's Hy3 large model sees 68x call volume growth in first week
Tencent's Hunyuan large model, Hy3, has seen a significant increase in total call volume, growing over 68 times compared to its predecessor, Hy2, within its first week of release. The Hy3 model has demonstrated a steeper growth trajectory and achieved the top ranking on OpenRout…
-
AssemblyAI tutorial shows dictation app with reduced latency
AssemblyAI has released a tutorial demonstrating how to build a dictation application called Blurt using their Sync API. The tutorial focuses on minimizing perceived latency by employing a connection pre-warm technique, aiming to achieve response times around 300ms. This approac…
-
AssemblyAI launches Sync API for faster, real-time audio transcription
AssemblyAI has introduced a new Sync API for audio transcription, designed for speed on short audio clips. Unlike the traditional async method where audio is processed in the background and results are collected later, the Sync API returns a transcript within the same HTTP respo…
-
Anthropic's Claude Fable 5 returns after export control outage amid developer concerns
Anthropic's Claude Fable 5, released on June 9, 2026, was temporarily taken offline due to U.S. export controls just three days later. The model was reinstated on July 1, but faced developer backlash concerning its safety filters and hidden routing mechanisms. The specific reaso…
-
Ex-OpenAI execs launch Inkling, an open-weight AI model
Thinking Machines Lab, founded by former OpenAI executives, has released its first open-weight model named Inkling. This model is designed to process audio, video, and text, featuring advanced reasoning and coding capabilities, though it's not top-ranked on all benchmarks. The c…
-
OpenAI releases GPT-5.6 Sol with record coding and cybersecurity scores
OpenAI has officially launched its GPT-5.6 Sol model family, featuring three tiers: Sol, Terra, and Luna. The flagship Sol model achieved a record 91.9% on the TerminalBench 2.1 coding benchmark and 96.7% on cybersecurity tests, prompting an initial government review due to its …
-
China approves 7 new mobile generative AI services, including Apple and Minimax
China's Cyberspace Administration has announced the approval of seven new generative AI services designed for mobile devices. This move is part of an effort to foster innovation and ensure the responsible application of AI technologies. Among the approved services are those from…
-
Phantom wallet enables AI agents to sign transactions across blockchains
Phantom has released a new MCP server that allows AI agents to directly operate a Phantom wallet, enabling them to sign messages and transactions across multiple blockchains like Solana, Ethereum, and Bitcoin. This infrastructure advancement allows agents to act as signers, clos…
-
OpenAI uses AI to find AI flaws, outperforming human testers
OpenAI is employing its internal GPT-Red model to identify vulnerabilities in its own AI systems, achieving an 84% success rate in simulated attacks. This AI-driven approach significantly outperforms human red teamers, who only managed a 13% success rate. The insights gained fro…
-
Unified API endpoint simplifies access to multiple LLMs including DeepSeek, Qwen, GLM
A new API service called aibridge-api.com offers a unified endpoint to access multiple large language models, including those from DeepSeek, Qwen, and General Language Model (GLM), alongside OpenAI-compatible models. This service simplifies integration by requiring only one API …
-
German consortium releases open hybrid Mamba-Transformer model Soofi S 30B-A3B
A German research consortium has released Soofi S 30B-A3B, an open-source, hybrid Mamba-Transformer Mixture-of-Experts foundation model. Trained on Deutsche Telekom's AI cloud using NVIDIA B200 GPUs, the model prioritizes German language data, significantly outperforming its arc…
-
AI agents gain persistent memory with new MCP server implementation
A technical article details the creation of an MCP Server designed to provide AI agents, specifically Claude, with persistent long-term memory. This server implements a memory system that reinforces frequently used facts and preferences while allowing less used information to de…
-
Apple CarPlay to support video playback with iOS 27 update
Apple is introducing a significant update to CarPlay with iOS 27, enabling video playback directly on vehicle screens when the car is stationary. This feature, announced at WWDC 2026, will allow users to browse and play videos from compatible apps, similar to services like Apple…
-
Databricks integrates Thinking Machines Lab's Inkling model for enterprise AI
Databricks has announced the integration of the Inkling model, developed by Thinking Machines Lab, into its Unity AI Gateway. This open-weights model is optimized for coding, agentic reasoning, and multimodal inputs, allowing enterprises to build and deploy AI agents and applica…
-
New Study Finds Up to 77% of AI Features May Be Non-Functional
A new study has revealed that a significant portion of features identified by Sparse Autoencoders (SAEs), a tool used in mechanistic interpretability, may not actually be functional. The research found that up to 77% of SAE features, despite passing standard cosine similarity me…