Summary
DeepSeek moved its V4 family out of preview into general availability on July 20, 2026, shipping V4-Pro (1.6T total / 49B active parameters) and V4-Flash (284B / 13B active), both with 1M-token context, and adding agentic capabilities plus stronger math and code generation. On July 24 it fully retired the legacy deepseek-chat and deepseek-reasoner aliases, forcing developers off hard-coded endpoints, and introduced peak/off-peak API pricing.
What changed
V4-Pro and V4-Flash reached GA on 2026-07-20 with agentic, math-reasoning, and code-generation gains over the preview builds. The legacy deepseek-chat and deepseek-reasoner aliases became inaccessible after 2026-07-24 15:59 UTC, and the API added a peak/off-peak pricing mechanism where listed prices double during peak hours.
Why it matters
GA plus a hard alias cutover pushes cautious enterprises to move production workloads onto V4, while peak/off-peak pricing is a notable cost-structure change for a widely used low-cost API. The timing clusters with GPT-5.6, Claude Opus 5, and Kimi K3, intensifying price and capability competition at the value end of the model market.
Evidence excerpt
On July 20, 2026, DeepSeek pushed its V4 model family out of preview into general availability; the legacy aliases deepseek-chat and deepseek-reasoner became inaccessible after July 24, 2026, 15:59 UTC.