DeepSeek V4-Pro and V4-Flash graduate from preview to official release in mid-July 2026 with a 1-million-token context window and new peak/off-peak API pricing.
By Quantara Editorial Team · AI Tools · Published · 7 min read
DeepSeek V4 is moving from preview to official release in mid-July 2026, closing out roughly three months of preview testing that began on April 24, 2026. The change is more about graduation and pricing than an all-new model, but it comes with a lineup-wide context upgrade and a notable pricing shift.
Two Models, One Lineup
| Model | Parameters | Positioning |
|---|---|---|
| DeepSeek-V4-Pro | 1.6T total / 49B active (MoE) | Performance positioned to rival top closed-source models |
| DeepSeek-V4-Flash | 284B total / 13B active (MoE) | Faster, more economical option |
Both models were first released as preview versions on April 24, 2026. The mid-July official release keeps the same core lineup while adding further performance upgrades and removing the preview label.
What's New at Official Release
The full lineup now standardizes on a 1-million-token context window, with reported improvements in agent-based task execution, mathematical reasoning, and code generation. That puts DeepSeek V4 in the same context-length tier as several other frontier-adjacent open and hosted models currently on the market.
Peak and Off-Peak API Pricing
DeepSeek introduced peak and off-peak pricing for the first time alongside the official release. Peak hours run 9:00am–12:00pm and 2:00pm–6:00pm daily, during which API usage is charged at 2x the off-peak rate. Teams running batch or non-urgent workloads can capture meaningful savings simply by scheduling jobs outside those windows.
What This Means for Teams Already on the Preview
- Re-check pricing assumptions in any cost model built during the preview period, since peak/off-peak billing is new.
- Consider shifting batch jobs, evaluations, and non-interactive workloads to off-peak hours.
- Re-test agentic and code-generation tasks against the official release, since DeepSeek reports further tuning beyond the preview build.
Sources
- TechNode: DeepSeek to launch V4 in mid-July with new peak-time API pricing
- SitePoint: DeepSeek V4 Released — What's New in the Latest Model
- DeepSeek API Docs: DeepSeek V4 Preview Release
Frequently asked questions
- What is the difference between DeepSeek V4-Pro and V4-Flash?
- V4-Pro is a 1.6T-total / 49B-active-parameter MoE model positioned to rival top closed-source models. V4-Flash is a smaller 284B-total / 13B-active-parameter model built for faster, more economical use.
- Is DeepSeek V4 a new model or an update to the preview?
- It is the official graduation of the preview models first released on April 24, 2026, with further performance improvements, rather than an entirely new model generation.
- What is DeepSeek's peak/off-peak pricing?
- Starting with the official V4 release, API usage during peak hours (9:00am–12:00pm and 2:00pm–6:00pm daily) is charged at 2x the off-peak rate. Scheduling non-urgent workloads outside those windows can reduce costs.
- What context window does DeepSeek V4 support?
- The full official lineup standardizes on a 1-million-token context window across both V4-Pro and V4-Flash.