DeepSeek V4 set for July with 1M-token context, peak pricing
1M-token context and time-based API pricing arrive in mid-July.
DeepSeek confirmed on Monday that its V4 model will see an official release in mid-July, building on the preview with enhanced performance and features. The headline upgrade is a 1-million-token context window now standard across the entire model lineup, enabling applications like long-document analysis, full-codebase reasoning, and extended conversation histories. In addition, DeepSeek V4 delivers measurable gains in agent-based workflows, math problem-solving, and code generation, making it competitive with frontier models from OpenAI and Anthropic.
Alongside the release, DeepSeek introduces its first dynamic pricing scheme: peak API usage hours (9:00 AM–12:00 PM and 2:00 PM–6:00 PM daily) will be charged at double the off-peak rate. This encourages developers to batch non-urgent workloads during off-peak times, reducing costs for latency-tolerant tasks. The move aligns with industry trends toward usage-based pricing (e.g., AWS Lambda’s tiered compute pricing) and gives teams a lever to optimize AI spend. For professionals, DeepSeek V4 combines massive context capacity with a flexible cost model, making it a strong candidate for enterprise RAG pipelines, coding assistants, and multi-step agent systems.
- Official release of DeepSeek V4 in mid-July with a 1M-token context window across all model variants.
- Stronger performance in agent tasks, math reasoning, and code generation compared to the preview.
- New peak/off-peak API pricing: double the off-peak rate during 9–12 AM and 2–6 PM daily.
Why It Matters
Developers gain a massive context window and can cut API costs by shifting workloads to off-peak hours.