DeepSeek V4 drops mid-July with 1M-token context and peak pricing

DeepSeek confirmed that version 4 of its large language model will officially launch in mid-July, building on the earlier preview. The release brings a unified 1-million-token context window to every model tier, enabling longer, more coherent interactions without truncation. Performance improvements are concentrated in three areas: agent-based task execution (multi-step tool calling and planning), mathematical reasoning (complex problem-solving accuracy), and code generation (syntax and logic fixes). The company says these enhancements stem from refinements to the underlying architecture and training pipeline.

Alongside the model upgrade, DeepSeek unveiled a new pricing structure that adds time-of-day surcharges for API usage — a first for the company. Peak hours run from 9:00 a.m. to 12:00 p.m. and 2:00 p.m. to 6:00 p.m. local time each day, during which API calls will be billed at twice the off-peak rate. The move is designed to manage server load and incentivize off-peak development. While exact per-token prices have not been released, the tiered pricing model marks a shift toward capacity-aware billing that could affect high-volume users who operate during business hours.