Frontier Models
DeepSeek V4 Enters General Availability with Peak API Pricing
The official launch introduces optimizations for V4-Pro and V4-Flash while retiring legacy endpoints and applying new pricing structures to the open-weight models.
DeepSeek V4 is a series of frontier AI models from DeepSeek that utilize mixture-of-experts architecture to deliver high performance with efficient parameter activation and one million token context support.
DeepSeek moved its V4 series from preview to official general availability in mid-July 2026. The release incorporates feature optimizations and performance improvements across the models. Peak-hour API pricing entered effect at the same time as the official launch.
What background led to the DeepSeek V4 official release?
DeepSeek launched the preview version of its V4 series on April 24, 2026. The preview made V4-Pro and V4-Flash available to users through the API and as open weights. Both models received immediate support for one million token context length by default. The preview announcement highlighted the cost-effective nature of the one million token context capability.
The preview established the mixture-of-experts design for the series. DeepSeek positioned the models as strong performers in the frontier category. Users accessed the models through the DeepSeek API during the preview period. The company indicated plans for an official version during the preview phase.
What technical specifications define the V4 models?
DeepSeek-V4-Pro contains 1.6 trillion total parameters with 49 billion active parameters per token. DeepSeek-V4-Flash contains 284 billion total parameters with 13 billion active parameters per token. The design allows efficient computation while maintaining high capacity through the mixture-of-experts structure.
| Model | Total Parameters | Active Parameters | Context Length | License |
|---|---|---|---|---|
| DeepSeek-V4-Pro | 1.6T | 49B | 1M tokens | MIT |
| DeepSeek-V4-Flash | 284B | 13B | 1M tokens | MIT |
The models operate under the MIT license. Open weights for both variants became available on the Hugging Face platform. This release method enables direct download and local deployment by developers and researchers.
What timeline applies to the API changes and legacy retirement?
- April 24, 2026: Preview launch of V4-Pro and V4-Flash models with one million token context.
- Mid-July 2026: Official general availability release with feature optimizations and peak-hour API pricing introduction.
- July 24, 2026 at 15:59 UTC: Discontinuation of legacy API aliases deepseek-chat and deepseek-reasoner.
The legacy aliases previously routed traffic to V4-Flash modes. Applications using the old names required updates before the July 24, 2026 cutoff. The change log from DeepSeek documented the three-month notice period starting from the April preview date.
What performance results does DeepSeek-V4-Pro show on benchmarks?
The 80.6 percent result applies specifically to the Max reasoning mode configuration. This benchmark measures software engineering task resolution. The score reflects the model's capability in coding-related evaluations.
The official version of DeepSeek V4 is planned to launch in mid-July.DeepSeek, Company
How does the release affect market economics for open-weight models?
The MIT license and open weights place the models in direct competition with other open frontier releases. Peak and off-peak API pricing creates differentiated cost structures for high-demand periods. The combination of large context length and efficient active parameters supports broader adoption in cost-sensitive applications.
Stakeholders in the AI industry gain additional options for high-context workloads. The retirement of legacy endpoints standardizes access through the current model names. Developers must adjust integration points to maintain continuity after the July 24, 2026 date.
What implications arise for users and the industry?
Users benefit from the one million token context in both Pro and Flash variants without additional configuration. The open release supports fine-tuning and modification by third parties. Performance improvements in the official version build on the preview foundation.
The discontinuation enforces migration to supported endpoints. Peak-hour pricing introduces planning requirements for API usage patterns. The overall release strengthens the position of open-weight models in the frontier segment.
What developments may follow the DeepSeek V4 launch?
Further iterations could incorporate additional optimizations based on user feedback from the official release. The community can contribute to the models through the open weights on Hugging Face. Continued competition in the open-weight space may influence pricing strategies across providers.
Frequently asked
When did the DeepSeek V4 preview launch?
The preview launched on April 24, 2026, introducing V4-Pro and V4-Flash with one million token context support.
When were the legacy API aliases discontinued?
The legacy API aliases deepseek-chat and deepseek-reasoner were discontinued on July 24, 2026, at 15:59 UTC.
What license governs the DeepSeek V4 models?
The models are released under the MIT license with open weights available on Hugging Face.
Sources
- DeepSeek — The two legacy API model names, deepseek-chat and deepseek-reasoner, will be discontinued in three months (2026-07-24).
- DeepSeek-AI — We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens. DeepSeek-V4-Pro resolves 80.6% of tasks on SWE-bench Verified in Max reasoning mode.
- DeepSeek — DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.
- ExplainX — The official version of DeepSeek V4 is planned to launch in mid-July.