Latest models
- DeepSeek V4.1-Flash (September 10, 2026) Current
A 552 billion parameter mixture-of-experts model with native visual understanding, available in the DeepSeek API as deepseek-flash. DeepSeek is phasing out V4-Pro. Source: DeepSeek API docs ↗
Checked October 7, 2026. See the full AI model tracker.
What we like
- V4.1-Flash is natively multimodal
- Lower API prices; DeepSeek says its cache needs far less memory than the previous generation
- Earlier V4 models were released openly
Watch out for
- Model names and routing change quickly, so read the changelog
- V4-Pro is being phased out and routed to V4.1-Flash
- Check data-handling and compliance requirements before using any hosted API
DeepSeek is a Chinese AI lab known for efficient, openly released models. Its V4 series was previewed on April 24, 2026.
What is current
DeepSeek-V4.1-Flash was announced on September 10, 2026. It is a 552 billion parameter mixture-of-experts model with a new encoder-decoder design (8 billion active parameters for input, 16 billion for output) and native visual understanding. In the API, set the model to deepseek-flash. DeepSeek says it is phasing out V4-Pro, and deepseek-v4-pro requests route to V4.1-Flash until V4.1-Pro launches. See our news summary and the model tracker.
Who should use it
Developers who want an inexpensive API model for coding and agent workflows. Review your model settings and costs after the V4-Pro change.
Compare assistants: ChatGPT, Claude and Gemini.