AI Briefing#deepseek-v4
The latest AI and developer news about #deepseek-v4, with the original source and a short summary.
The latest AI and developer news about #deepseek-v4, with the original source and a short summary.
A ByteDance Seed paper reveals that DeepSeek V4's chunked KV-cache compression creates periodic retrieval failures, with base model accuracy varying by 40.2 percentage points depending on token position.
vLLM v0.28.0 adds large-scale performance optimizations and inference capabilities for the Kimi-K3 and DeepSeek V4 models.
vLLM v0.28.0 introduces major performance optimizations and new inference features for Kimi-K3 and DeepSeek V4.
DeepSeek has released the V4-Flash API in public beta with enhanced agent performance.
NVIDIA ModelExpress dramatically reduces model weight loading times through direct GPU-to-GPU data transfer.
NVIDIA is dramatically lowering the cost per token in AI inference through its Blackwell platform and optimized software stack.
A lightweight retriever that compresses DeepSeek-V4's CSA KV-cache to maximize GPU memory efficiency has been released.
The latest open-weight LLMs have adopted KV cache compression and attention optimization techniques for long-context efficiency
Open-weight LLMs including Gemma 4 and DeepSeek V4 are lowering long-context costs through KV sharing and compressed attention.
The official DeepSeek V4 paper has been released, adding FP4 QAT and stabilization techniques.
CAISI evaluated DeepSeek V4 and found it to be China's top model, but about 8 months behind the US frontier.
DeepSeek released previews of V4-Pro and V4-Flash, emphasizing price competitiveness.
An analysis of DeepSeek-V4's architectural innovations, infrastructure optimizations, and the trial-and-error process during training.
It was confirmed that the DeepSeek V4 API has a technical requirement to always pass back reasoning_content when Tool Calls are involved.
With DeepSeek V4 going open-source, China's AI industry is restructuring away from model development and toward infrastructure and specialized data.
NVIDIA unveiled Blackwell Day-0 support for DeepSeek V4 with performance of 3,500TPS per second.
DeepSeek has released the V4 model family, which supports a 1M token context.
DeepSeek unveiled two 1M token context V4 preview models, pushing a low-price strategy.
DeepSeek-V4 was released, featuring 1M-token context and strong coding benchmark results.
A DeepSeek-V4 preview has been released, supporting a 1.6T MoE and 1M context.
DeepSeek has unveiled two variant models, **V4-Pro** and **V4-Flash**.
DeepSeek's latest models, the V4 Flash and Non-Flash versions, have been released via Hugging Face.
DeepSeek-V4 unveiled a 1M-token context and agent optimizations.