DeepSeek V4 Stable Release Transitions from Preview; Legacy API IDs Retire July 24

DeepSeek announced that legacy API model names deepseek-chat and deepseek-reasoner will be fully retired on July 24, 2026 at 15:59 UTC, completing the transition from the V4 Preview (launched April 24, 2026) to stable production. The two flagship models, deepseek-v4-pro (1.6T total / 49B active parameters) and deepseek-v4-flash (284B total / 13B active parameters), both ship with 1M-token context, MIT-licensed open weights on Hugging Face, and official API access. The V4 Preview introduced a hybrid Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) architecture targeting inference efficiency; V4-Pro scores 80.6% on SWE-bench Verified (tied with Claude Opus 4.6 at 80.8% within margin).

DeepSeek has announced peak-hour pricing for the V4 stable release: 2× baseline cost during Beijing business hours (9-12 and 14-18 UTC+8). Off-peak rates remain at current preview levels. V4-Pro list price is $1.74 / $3.48 per 1M input/output tokens, though a 75% discount brought it to $0.435 / $0.87 through May 2026. V4-Flash remains roughly 1/10th the output cost of V3.2 at $0.14 / $0.28. Developers using the deprecated legacy IDs have until July 24 to migrate; until then both aliases route to V4-Flash in non-thinking and thinking modes respectively, making migration a single-line code change.

For production AI teams, the stable release marks DeepSeek V4's graduation from optional to locked production status. The efficiency-first positioning—90% frontier capability at 40-50% of rival API costs—has gained traction as enterprises prioritize deployment cost over marginal capability gains. V4-Flash has become a default agent backbone for cost-sensitive workloads; V4-Pro targets code generation and reasoning tasks where the 1.6T parameter scale justifies cost trade-offs. The July 24 deadline creates a soft forcing function for migration planning, though existing integrations continue to work with legacy IDs until the cutoff.