SenseNova 6.7 Flash-Lite by SenseTime Reduces Consumption by 60%

SenseTime has officially launched its new lightweight multi-modal intelligent agent model — SenseNova 6.7-Lite. Designed to meet the needs of "real-world streams," the model uses a multi-modal architecture to directly understand complex layouts, document structures, and financial charts, achieving an integrated "see, think, act" process and improving the success rate of tasks such as data analysis, in-depth research, and PPT generation.

Technically, SenseNova 6.7 Flash-Lite eliminates the intermediate visual layer, achieving a significant leap in agent capabilities with fewer parameters. This innovation has led to numerous SOTA (State Of The Art) results in authoritative benchmark tests. In addition, the model significantly reduces token consumption during the reasoning process, decreasing by 60% compared to traditional text-only agents. At the same time, SenseNova 6.7 Flash-Lite can provide millisecond-level feedback, meeting the needs of high-frequency interactions.

Currently, this model is widely applied across multiple industries such as finance, manufacturing, healthcare, and education, and it possesses five core capabilities, including action decision-making, autonomous arrangement of tool chains, perception in noisy environments, self-correction and recovery, and long-term memory consistency. These capabilities bring higher reliability and execution stability to real-world workflow scenarios, redefining the standards for lightweight multi-modal intelligent agent models.

To encourage developers to use this model, SenseTime has also announced a limited-time open SenseNova Token Plan, offering free quotas in the first month where developers can refresh 1500 call quotas every 5 hours. Meanwhile, the entire SenseNova-Skills line of office skills is also open-sourced, providing developers with more tools and resources.

Key Points:
🌟 SenseNova 6.7 Flash-Lite adopts a lightweight design, enhancing the capabilities of multi-modal agents.
⚡ Token consumption has been reduced by 60% compared to traditional models, meeting the needs of high-frequency interactions.
💼 The model supports multiple industries and offers a limited-time free usage plan to help developers.

The ChatGPT Moment in the Robot World? Genesis AI Completes Egg Beating and Piano Playing with a Single Model. This Demo is Amazing!

Startup team Genesis AI releases its first robot foundation model, GENE-26.5, marking a breakthrough in general-purpose robots for high-difficulty unstructured tasks. In demonstrations, the robot exhibits remarkable flexibility and autonomy, performing delicate tasks such as single-handedly cracking eggs and solving Rubik's cubes, enhancing control over force, logic, spatial perception, and rhythm precision.....

Apple to Introduce Multimodel Selection Mechanism in iOS 27, Supporting Third-Party AI Access to System Functions

Apple is advancing its AI strategy by introducing a multi-model selection mechanism codenamed 'Extensions' in iOS 27, iPadOS 27, and macOS 27. Users can on-demand invoke various large language models for system functions like Siri, writing tools, and image generation. This marks Apple's shift from a single AI provider to a platform-based model orchestration and distribution architecture.....

OpenAI System Prompt Leaked, New Model GPT-5.5 Strictly Prohibits Discussion of Goblins

OpenAI's latest open-source Codex CLI accidentally exposed GPT-5.5's system prompts, including a mysterious instruction prohibiting discussions of fantasy creatures like 'goblins' and 'elves' in conversations. The 3,500-word base instruction set mandates that the model avoid these topics unless user queries have absolutely clear relevance, aiming to prevent AI from falling into specific hallucinations.....

Anthropic to Start Massive Fundraising with a Valuation of $90 Billion, Possibly Surpassing OpenAI Before IPO

AI unicorn Anthropic, targeting an IPO, has initiated its final private funding round, securing approximately $50 billion in financing offers at a valuation between $850 billion and $900 billion. With annualized revenue surpassing $30 billion and exponential growth, investor competition underscores strong market confidence in its technological potential.....

Solving the Memory Issues of Large Models: Huawei Expert Starts a Startup in Shenzhen, MemoraX AI Secures $10 Million in Funding

Amid the rapid development of AI large models, the issue of memory limitations has become increasingly prominent. Shenzhen-based startup MemoraX AI focuses on solving this pain point, recently completing a multi-million dollar seed round led by L2F Light Source Entrepreneur Fund and Zhongding Capital, with participation from several other institutions. Located in Nanshan, Shenzhen, the company has achieved remarkable speed from inception to fundi....