Realtime AI News
DeepSeek registers DeepSeek-V4-Flash-0731 on Hugging Face, its newest V4 Flash update
DeepSeek's official Hugging Face registry added DeepSeek-V4-Flash-0731 on July 31, a text-generation model built on transformers with safetensors weights and 8-bit support. It ships under the MIT license with endpoint-compatible tags, and the fresh entry had zero downloads at listing time.
DeepSeek added DeepSeek-V4-Flash-0731 to its official Hugging Face registry on July 31, the latest official update to the V4 Flash line, with the version stamped by a date suffix.
The registry metadata shows the model targets text-generation, is built on the transformers library, stores weights in safetensors format, and carries the deepseek_v4 tag that places it squarely in the V4 family.
It is released under the MIT license, leaving room for commercial use and downstream development, and is tagged endpoints_compatible and 8-bit, signaling support for standard inference endpoints and 8-bit quantized deployment.
The entry also links to an arXiv paper numbered 2606.19348, but this update is itself an official model-registry release; at the time of listing the entry showed zero downloads and two likes, marking it as a brand-new version.
The dated refresh continues DeepSeek's rapid open-source cadence: shipping new versions with date suffixes at high frequency hands the latest training output directly to the community and developers for testing.
For teams that deploy open models, the MIT license combined with endpoint compatibility means the new version can be wired into existing inference stacks at low cost, while 8-bit support lowers the memory bar for local runs.
The things to watch next are real-world evaluation results for this build and whether it appears on DeepSeek's official API; community reception will decide whether it replaces the previous Flash release as the default choice.
Why it matters
The date-stamped refresh keeps DeepSeek's open-source supply cadence moving, and the MIT license plus endpoint compatibility let developers adopt the build at near-zero cost, reinforcing DeepSeek's position in the deployment ecosystem.
Nearby Updates
All07/31, 20:00
Repeat founder Ryan Williams raises $10M seed for an AI startup for private credit managers
TechCrunch reports that repeat founder Ryan Williams has raised a $10 million seed round for an AI startup serving private credit managers. The company plans to bring AI automation to credit workflows in one of asset management's fastest-growing segments.
07/31, 21:49
Kimi K3 Becomes the Focus of Silicon Valley as Moonshot AI's Valuation Tops $35 Billion
Reports say Moonshot AI's valuation has surpassed $35 billion, with its flagship model Kimi K3 drawing intense attention in Silicon Valley. The figure underscores the rising global standing of Chinese AI labs in the current race.
07/31, 22:47
Smallest.ai raises $13M to build ultra-fast, human-sounding voice AI
Voice AI startup Smallest.ai has raised $13 million to build ultra-fast voice models that sound genuinely human, TechCrunch reported. The company says its core goal is to make AI-powered phone calls pass the Turing test, so listeners cannot easily tell a machine is on the line.
07/31, 22:53
Tencent kills Story of Seasons Mobile as AI spending ends Japan partnership
Tencent is shutting down Story of Seasons Mobile, a decision that also ends its Japan partnership, according to Tech Times. The move is tied to the company's heavy AI spending, which is pushing it to re-prioritize non-core business lines.