Skip to main content
AI RADARModels
253•07/31/2026, 11:45•2 min read

DeepSeek Updates V4 Flash in API

DeepSeek Updates V4 Flash in API
Executive TL;DR30s read

DeepSeek has introduced a new version of V4 Flash that outperforms previous models in performance benchmarks. The model is available via API and promises new capabilities in the near future.

DeepSeek Updates V4 Flash in API

The new V4-Flash-0731 outperforms GLM 5.2 and V4 Pro Preview in benchmarks. It is still a model with 284B parameters, with 13B active. The model uses approximately the same number of tokens as the previous version, according to tests by Artificial Analysis. This makes the model significantly cheaper per task than GPT 5.6 Luna, even considering the new prices, but Luna solves tasks much faster and uses fewer tokens.

The new model operates through the Responses API and is fully compatible with Codex. Additionally, DeepSeek promises to release its own agent soon.

Weights have not been released yet. The model is currently only available in the API, while the old version remains in the chat. An update to V4 Pro is expected to be released in early August.

Why it matters

Analysis

This is important as the new DeepSeek model offers improved performance and efficiency, which can significantly impact AI solution development. It also demonstrates the growing competition in the AI model market.

Discuss in community

Share your questions and insights with developers

Join Telegram
+5 Points
← Back to news feed

Related in Ecosystem

Practical guides, core terminology, and agent skills