AI daily
2026-09-02
Selected
10
Sources
10
Featured
6
More
4
World Labs, founded by Fei-Fei Li, has released Atlas, a new multimodal world model pretrained from scratch to handle text, images, video, and 3D data. Atlas enables world generation, reconstruction, and simulation, and is currently available to select partners for early access.
SynthePulse InsightRead deep analysisGoogle has introduced Gemini 3.8 Flash Cyber, an AI model tailored for defensive security operations to automate the generation of reliable code patches. Demonstrating superior performance, it produces 2.6 times more correct patches than leading commercial models and is now available via priority access for government and critical infrastructure entities.
SynthePulse InsightRead deep analysisAlibaba released Qwen3.8-Max-0902, a new version of its AI model, with enhanced post-training for coding and professional work. It achieved the top score of 1691 on CodeArena's front-end programming leaderboard and offers a competitive price of $5 per million tokens. The model is now available via API on the Qwen AI platform and integrated into Qwen Office, Qoder, and Qwen APP.
SynthePulse InsightRead deep analysisInworld has officially released its Realtime TTS-2 text-to-speech model for general availability. Currently ranking #1 on Artificial Analysis' Controlled Voice Arena, it is touted as the fastest in its class, with improvements driven by user data from its earlier research preview phase.
SynthePulse InsightRead deep analysisResearchers from Renmin University and Alibaba introduced LoD at EMNLP'26, a new framework for detecting jailbreak attacks in Large Vision-Language Models. By learning internal safety patterns from benign inputs rather than relying on known attack data, LoD identifies unseen attacks as anomalies. The method demonstrated superior performance across multiple LVLMs compared to existing baselines.
VAST has released Tripo P2.0, a new AI 3D model that generates native quad topology directly, addressing the issue of editability and animation readiness in AI-generated models. The company also announced the completion of Series B and B+ funding totaling approximately 3 billion RMB (about $420 million), bringing cumulative funding to around 5 billion RMB and setting a record in the 3D-native intelligence sector.
SynthePulse InsightMembersRead deep analysisDingxi Zhichuang (Rhinovate) has completed a Pre-A funding round of several hundred million yuan, led by Dinghui Baifu with participation from multiple investors. The funds will be used to develop its materials science foundation model and autonomous laboratory systems, aiming to integrate AI scientists with physical experiments to speed up materials discovery and validation.
SynthePulse InsightMembersRead deep analysisSovereign robotic defense startup Aitan has exited stealth mode following a $41 million funding round co-led by Deep33 and Dell Technologies Capital, raising its total capitalization to $54 million. The company specializes in AI-powered robotic defense systems and edge autonomy software, leveraging over 500,000 combat hours gained with the Israel Defense Forces. The new funds will drive international expansion and the deployment of its 'Robotic Sovereignty-as-a-Service' platform for allied nations.
SynthePulse InsightRead deep analysisX Square Robot has introduced TwinDEX, a paired wearable and robotic manipulation system designed to collect training data without using a robot and transfer those demonstrations to real-world dexterous manipulation. The system uses matched three-finger nine-degree-of-freedom hardware and reports up to 5.3 times the effective data-collection throughput of on-robot teleoperation.
SynthePulse InsightRead deep analysisAfterQuery, an AI training data company, is raising a new round at a $3.2 billion valuation just 18 months after its founding, a tenfold increase in five months. The company specializes in collecting expert process data from finance, law, and medicine to train reasoning models and agents. This funding round underscores the growing value of expert human data in AI development and the sustained demand for high-quality training data.