DeepSeek V4 preview goes live: 1M context, Pro/Flash variants, API available today
DeepSeek's April 24 official release makes V4 the new flagship family with 1M context, named Pro and Flash variants, and pricing that has since been cut further. V4-Pro is now listed at $0.435/$0.87 per 1M input/output tokens, with those rates becoming the official quarter-price baseline after the 75% discount window ends on May 31, 2026.
Official release snapshot
DeepSeek has officially released DeepSeek V4 and positioned it as the new flagship family for coding, long-context reasoning, and agent workflows.
What is officially confirmed
- 1M-token context window
- Two production model routes:
deepseek-v4-proanddeepseek-v4-flash - OpenAI-compatible migration path using the same endpoint shape
- Thinking / Non-Thinking modes for both official V4 variants
Practical meaning for developers
This release matters because it turns DeepSeek V4 from rumor-cycle discussion into a concrete deployment target. Teams can now test a current DeepSeek flagship without rebuilding their client stack: keep the endpoint pattern, switch the model ID, validate quality, then expand routing gradually.
Practical pricing correction
DeepSeek V4 Pro is now listed at $0.435 per 1M cache-miss input tokens and $0.87 per 1M output tokens, with cache-hit input at $0.003625 per 1M tokens (RMB 0.025 on the Chinese price table). DeepSeek's current pricing note says these V4-Pro rates will be officially adjusted to 1/4 of the original price after the 75% discount window ends on May 31, 2026.
DeepSeek V4 Flash is the route to watch for high-volume production traffic at $0.14 per 1M input tokens and $0.28 per 1M output tokens, with cache-hit input at $0.0028 per 1M. Its quality is excellent for everyday coding, chat, retrieval, and repeated tool steps, while preserving the same 1M-context headline.
Routing guidance
- Start with
deepseek-v4-flashfor high-volume chat, tool steps, and everyday coding loops. - Escalate to
deepseek-v4-profor harder reasoning, review-heavy coding, and longer evidence chains.
Legacy alias note
DeepSeek has also published a retirement path for older alias names such as deepseek-chat and deepseek-reasoner. New integrations should target the V4 model IDs directly.
Why this page exists
This hub treats DeepSeek as the headline model. The official V4 release strengthens that positioning because the product story is now clear: current flagship, named variants, long context, direct migration guidance, and a sharper Flash cost story for 1M-context workloads.