DeepSeek Releases V4-Flash and V4-Pro Models as Tencent Ships Hy3-Preview
DeepSeek has released two new models in its V4 series: DeepSeek-V4-Flash and DeepSeek-V4-Pro, both now available on Hugging Face. Separately, Tencent has shipped Hy3-Preview, marking simultaneous releases from two major Chinese AI labs.
DeepSeek Releases V4-Flash and V4-Pro Models as Tencent Ships Hy3-Preview
DeepSeek has released two new language models—DeepSeek-V4-Flash and DeepSeek-V4-Pro—both now available on Hugging Face. The releases coincide with Tencent's launch of Hy3-Preview, marking a concentrated wave of new model deployments from Chinese AI labs.
DeepSeek V4 Series
The V4-Flash and V4-Pro models represent DeepSeek's latest iteration following its V3 release. Based on Hugging Face repository metadata, both models have been published under the deepseek-ai organization, though technical specifications including parameter count, context window size, and benchmark performance remain undisclosed at publication time.
The naming convention suggests V4-Flash is optimized for speed and efficiency, while V4-Pro targets higher performance workloads—a pattern consistent with other model families that separate latency-optimized and capability-optimized variants.
Model weights and configuration files are available through Hugging Face's model hub, indicating open weight access rather than API-only availability. Pricing information has not been disclosed.
Tencent Hy3-Preview
Tencent's Hy3-Preview model has also appeared on Hugging Face under the tencent organization. The "Preview" designation indicates this is likely an early access or experimental release rather than a production-ready model.
Detailed specifications for Hy3-Preview—including architecture details, training data cutoff, and performance metrics—have not been published in the model repository at this time.
Release Context
The simultaneous availability of three models from two separate organizations suggests coordinated or coincidental timing in the Chinese AI development ecosystem. DeepSeek has previously released competitive models including DeepSeek-V3, which achieved strong performance on standard benchmarks while maintaining relatively low inference costs.
All three models are accessible through Hugging Face's infrastructure, enabling researchers and developers to download weights directly rather than relying exclusively on API access.
What This Means
These releases continue the trend of Chinese AI labs publishing open-weight models, contrasting with the API-only approach favored by some Western labs. The lack of immediate technical documentation suggests these may be early releases with details to follow. Developers interested in testing these models can access them through Hugging Face, though production deployment decisions should await published benchmarks and pricing information. The V4-Flash naming specifically indicates DeepSeek is pursuing a multi-tier model strategy similar to Anthropic's Claude family or OpenAI's GPT-4 variants.
Related Articles
Qwen 3.6 27B Released With FP8 Quantization, OpenAI Deploys Privacy Filter Model
Alibaba Cloud released Qwen 3.6 27B, a 27-billion parameter language model, alongside an FP8 quantized version for deployment efficiency. Separately, OpenAI published a privacy filter model on Hugging Face, marking a rare public model release from the company.
Tencent releases HY-OmniWeaving multimodal model as Gemma-4 variants emerge
Tencent has released HY-OmniWeaving, a new multimodal model available on Hugging Face. Concurrently, NVIDIA and Unsloth have published optimized variants of Gemma-4, including a 31B instruction-tuned version and quantized GGUF format.
Google Launches Native Gemini App for Mac, Bringing AI Assistant to Desktop
Google released a native Gemini application for macOS, marking the company's first standalone desktop client for its AI assistant. The app brings Gemini functionality directly to Mac users without requiring a web browser.
Mozilla finds 271 vulnerabilities in Firefox 150 using Anthropic's Claude Mythos Preview
Mozilla's Firefox engineering team identified 271 vulnerabilities for version 150 using Anthropic's Claude Mythos Preview, following a prior collaboration that yielded 22 security-sensitive fixes in version 148 using Opus 4.6. The findings demonstrate that AI models can now match elite human security researchers at discovering code vulnerabilities.
Comments
Loading...