ainewsblitz.com

Breaking

DeepSeek Releases V4-Flash With MIT-Licensed Weights, Touting Sevenfold Benchmark Jump Over Preview

DeepSeek Releases V4-Flash With MIT-Licensed Weights, Touting Sevenfold Benchmark Jump Over Preview
  • Foundation Models
  • Open Source
  • Software Dev & Coding

DeepSeek has released the production version of DeepSeek-V4-Flash, a low-cost mixture-of-experts model whose agentic coding scores leapt roughly sevenfold over its preview build, reigniting debate over how such gains are measured. The company published the model, dated July 31, 2026, on Hugging Face under a permissive MIT license and made it available through its official API.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,934 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year