AI Briefing#deepseek
The latest AI and developer news about #deepseek, with the original source and a short summary.
The latest AI and developer news about #deepseek, with the original source and a short summary.
Chinese AI startup DeepSeek has begun developing inference chips for its own data centers in response to US export restrictions.
Distillation, which began as a model compression technique, has now become a core technique for transferring reasoning ability between models and the center of a copyright controversy.
A proxy attack exploiting Claude Desktop's MCP connector permissions and a DeepSeek-based ransomware PoC have been reported.
DeepSeek has released DSpark, a speculative decoding framework that dramatically improves LLM response speed, under the MIT license.
A Pull Request to support the DeepSeek V4 model has been merged into the llama.cpp project.
DeepSeek is introducing a peak-valley pricing system starting mid-July that charges 2x the standard price during certain hours.
DeepSeek released DeepSpec, a tool for training and evaluating models for Speculative Decoding.
DeepSeek has open-sourced an optimization technique that maximizes inference efficiency, boosting generation speed by up to 85%.
Chinese AI startup DeepSeek secured its first outside funding at a $50 billion valuation, on the condition that investors refrain from recruiting its core talent.
The US government has held off on adding over 100 Chinese AI and semiconductor companies, including DeepSeek, to its trade blacklist.
China's Weibo has released a 3B-scale VibeThinker model that has become controversial after posting benchmark scores that surpass ultra-large models.
The strengthening capabilities of DeepSeek researchers and the growing independence of China's domestic AI talent-cultivation ecosystem have emerged as a policy challenge for the United States.
Vercel's AI Gateway has started supporting DeepSeek V4 Pro and Flash models via Azure.
A record of an on-site visit regarding DeepSeek's corporate operating style and the Chinese AI industry ecosystem.
DeepSeek V4 Pro won by showing higher accuracy and instruction-following ability than GPT-5.5 Pro across 4 precision task tests.
DeepSeek is rapidly expanding its token share on the back of low prices, while Anthropic continues to lead in actual AI spending volume despite its high costs.
Introducing On-policy distillation (OPD), a core training technique used in the latest models such as Qwen and DeepSeek.
Dropstone 1.5, an AI coding agent that uses the latest open models at a lower price than Claude Code, has been released.
Tencent's Hy3 model has surpassed Claude to take an overwhelming No. 1 spot on OpenRouter's usage rankings.
To prevent technology leaks, the Chinese government is requiring key personnel at major AI companies to obtain government approval before traveling abroad.
As frontier model costs rise, combining engineers from low-wage countries with local AI is emerging as a more economical alternative.
Chinese AI startup DeepSeek has permanently cut the API price of its flagship model V4-Pro by 75%.
It is an open-source terminal coding agent that dramatically lowers costs by optimizing prefix caching on the DeepSeek API.
Introducing DeepSeek reasonix, a coding agent that leverages DeepSeek models to boost caching efficiency and cut costs.