# DeepSeek V4 is out, bringing major optimizations, including up to 1.6T model sizes.

*semiconductor · news · 2026-04-26 · Wccftech*

## Key points

- DeepSeek V4 introduces a 1.6 trillion parameter Pro model and a 284 billion parameter Flash version.
- V4 reduces single-token inference FLOPs to 27% and KV cache to 10% for a 1M-token context.
- NVIDIA Blackwell GPUs provide Day-0 support for DeepSeek V4, achieving nearly 3500 TPS per GPU.
- DeepSeek V4 uses FP4 (MXFP4) quantization for faster rollouts and lower memory traffic and latency.
- Huawei's 2026 Ascend 950PR and 950DT chips will support MXFP4, ensuring DeepSeek V4 compatibility.

**Companies:** NVIDIA
**Countries:** China, United States

[Read the full story on Wccftech](https://wccftech.com/nvidia-beats-everyone-to-deepseek-v4-day-0-blackwell-support-pushing-3500-tokens-on-1-6t-models/)

---

Canonical: https://newsio.io/n/2d5d8692-d66a-4ab9-9bc9-9ef459c50239/deepseek-v4-is-out-bringing-major-optimizations-including-up-to-1-6t-model-sizes
Summarized by Newsio from Wccftech. https://newsio.io/how-it-works