Search AI Products and News

Beijing's Artificial Intelligence Industry Investment Fund has announced an additional investment of 200 million RMB in Zhipu (Z.ai), building on its previous investment, to support the development of Zhipu's open-source models and the construction of its open-source community ecosystem. Zhipu is the first AI large language model company invested in by the fund and is currently the fastest-growing company. Zhipu has comprehensive accumulation in model capabilities including text, reasoning, speech, image, video, and code, a well-established commercial layout, and a developer community and enterprise user base exceeding one million.

The Beijing Artificial Intelligence Industry Investment Fund recently announced an additional investment of 200 million RMB in Beijing Zhipu AI Technology Co., Ltd. (Zhipu). According to the fund, Zhipu is the first AI large language model company invested in since the fund's establishment and is one of the fastest-growing companies in the field. Zhipu has accumulated deep expertise in model development across text, reasoning, speech, image, video, and code, and boasts a robust commercialization strategy, having established a developer community and enterprise user base exceeding one million. This investment aims to further support Zhipu's growth and development.

Zhipu AI's technology team has announced the open-sourcing of its 32B and 9B GLM (General Language Model) model series, and the official launch of its new interactive platform, Z.ai. This model series includes base models, inference models, and contemplative models, all under a permissive MIT license. This grants developers extensive freedom for use and development, allowing free use for commercial purposes and free distribution.

Recently, the AI code generation field has seen a surge in open-source releases, with several heavyweight models making their debut. Among them, the Cogito v1 Preview series from Deep Cogito stands out. According to AIbase, this new family of open-source models includes various sizes: 3B, 8B, 14B, 32B, and 70B parameters. Not only does it outperform competitors in its class, but its 70B version even surpasses Meta's recently released Llama 4 109B MoE model, sparking considerable industry discussion.

On April 8th, 2025, NVIDIA launched Llama 3.1 Nemotron Ultra 253B, an open-source model optimized from Llama-3.1-405B. With 25.3 billion parameters, it surpasses Meta's Llama 4 Behemoth and Maverick, becoming a focal point in the AI field. This model demonstrates superior performance in benchmarks such as GPQA-Diamond, AIME 2024/25, and LiveCodeBench, achieving inference throughput comparable to DeepSeek.

OpenAI CEO Sam Altman announced on social media today that the company will release an open-source artificial intelligence model in the coming months. This news has generated significant attention, especially after the breakthrough success of the R1 model by Chinese company DeepSeek, putting pressure on OpenAI and highlighting competition from Meta's Llama series of models. Altman stated that OpenAI has long considered launching an open-source model.

Data intelligence company Databricks recently unveiled a novel large language model fine-tuning method – TAO (Test-time Adaptive Optimization). This technology offers new hope for the advancement of open-source models. By leveraging unlabeled data and reinforcement learning, TAO excels in reducing enterprise costs while achieving remarkable results across various benchmark tests. According to tech news outlet NeoWin, TAO...

On March 19th, an open-source text-to-speech (TTS) model called Orpheus TTS was officially launched. This model has quickly gained attention for its human-like emotional expression, natural and fluent voice quality, and ultra-low latency real-time output stream. Orpheus TTS reportedly excels in real-time conversational scenarios and promises to bring new breakthroughs to intelligent voice interaction. Orpheus TTS focuses on low latency and high emotional expression, with core features including: - **Ultra-Low Latency**: Default latency approximately 2

The AI revolution is upon us! Tech leader Kai-Fu Lee recently made a bold prediction: 2025 will not be ordinary; it will mark an explosive year for AI applications and a crucial test for large model commercialization. Opportunities and challenges coexist as a transformative storm reshaping the future of AI is brewing. Lee's prediction is not unfounded but based on his deep understanding of industry trends. He points to the emergence of DeepSeek as an example, expressing excitement that it represents more than just a Chinese AI advancement.

The Allen Institute for AI (AI2) recently unveiled its latest large language model, OLMo2 32B. This model boasts a significant advantage: it's fully open-source, directly challenging proprietary models. The most notable feature of OLMo2 32B is its completely open nature. AI2 has generously released all data, code, weights, and detailed training processes. This transparency sets a new standard.

Attention all visual enthusiasts seeking high-definition images! The image processing world welcomes a powerful new contender: Thera, an open-source arbitrary scale super-resolution model. This cutting-edge model offers a user-friendly open-source approach to significantly improve image clarity. Thera not only restores blurry photos to their former glory but also provides unparalleled flexibility in scaling images, making it a boon for detail-oriented users and a nightmare for blurry image enthusiasts! Want to enlarge an image to any arbitrary size? Frustrated with fixed magnification factors? Thera offers a solution, eliminating these limitations.

Recently, the 360 ZhiNao team announced the successful replication of DeepSeek's reinforcement learning results and the official release of the open-source reasoning model Light-R1-14B-DS. This model surpasses DeepSeek-R1-Distill-Llama-70B and DeepSeek-R1-Distill-Qwen-32B in performance, becoming the industry's first 14B parameter model to achieve reinforcement learning effects. It significantly improves mathematical reasoning capabilities, outperforming most 32B-parameter models.

Beijing Zhipu AI announced the release of CogView4, the first open-source text-to-image model capable of generating Chinese characters. The model achieved the highest overall score on the DPG-Bench benchmark, making it the State-of-the-Art (SOTA) in open-source text-to-image models. Furthermore, it's the first image generation model to be released under the Apache 2.0 license.

On February 25th, Alibaba announced the launch of QwQ-Max-Preview, a reasoning model based on Qwen2.5-Max, and plans to fully open-source its latest reasoning models, QwQ-Max and Qwen2.5-Max. The released QwQ-Max-Preview is a preview version. Alibaba stated that the official version will be released soon and fully open-sourced under the Apache2.0 license. Unlike previous releases, this open-sourcing includes not only the model itself but also smaller versions such as Q...

In the age of AI, large language models (LLMs) are akin to martial arts secrets; their training processes consume massive computing power and data, much like martial arts experts who have spent years in secluded training. The release of open-source models is like these experts sharing their secrets with the public, but they come with certain licenses (such as Apache 2.0 and LLaMA 2 Community License) to protect their intellectual property (IP). However, the world can be treacherous, and incidents of 'shelling' are bound to occur. Some developers claim they have trained new LLMs when, in fact, they are using other foundational models (like Lla

In the field of game development, large models are becoming an indispensable 'think tank', showcasing comprehensive capabilities from AI character generation to scene construction. However, improvements are still needed in game scene understanding, image recognition, and content description. A research team from Alberta, Canada has launched the open-source large model VideoGameBunny (VGB), optimized for game development, featuring multi-language support, high customizability, powerful text generation capabilities, easy accessibility, compatibility with various development environments, a wealth of model files, and active community support. VGB supports multiple...