DeepSeek V4 Official Launch Set for Mid-July: Peak Pricing Changes Everything

DeepSeek confirmed on June 30 that the official version of DeepSeek V4 will launch in mid-July 2026, graduating from its current preview status with performance improvements and a first-of-its-kind peak-hour pricing model. For teams running production workloads on DeepSeek's API, the clock is about to become part of the cost model.

What's New in V4 Official

The official release builds on the V4 preview with stated improvements in three areas:

• Agent-based task execution — better multi-step tool use and planning
• Mathematical reasoning — improved accuracy on complex computation
• Code generation — stronger performance across languages

All V4 models ship with a 1-million-token context window standard across the lineup. The V4 family includes two variants:

• DeepSeek V4 Pro — full capability, higher pricing
• DeepSeek V4 Flash — optimized for speed and lower cost

Peak-Hour Pricing: The Big Change

DeepSeek is introducing time-based API pricing for the first time. Here is how it works:

Peak Hours (2x rate)

• 9:00 AM – 12:00 PM Beijing Time (UTC+8)
• 2:00 PM – 6:00 PM Beijing Time (UTC+8)

Off-Peak Hours (standard rate)

• All other hours, weekends, and holidays

During peak windows, both input and output token costs double across both V4 Pro and V4 Flash.

Timezone Translation

For US and European developers, DeepSeek's peak hours largely fall during overnight windows — meaning most Western API traffic hits off-peak rates. This is likely intentional. DeepSeek's heaviest usage is in China, and the peak pricing targets domestic business-hour congestion.

How This Compares

DeepSeek's peak pricing is the first time a major LLM provider has introduced time-based metering. Here is how V4 fits into the current landscape:

Even at peak pricing, DeepSeek V4 Pro at $1.00/M input and $4.00/M output undercuts every Western frontier model by a wide margin. The value proposition for cost-sensitive workloads remains strong.

Migration from V4 Preview

DeepSeek has indicated the migration path is straightforward:

• Preview API keys continue to work
• Model identifiers update from deepseek-v4-preview to deepseek-v4-pro and deepseek-v4-flash
• Preview pricing applies until the official launch date
• A migration guide is available in DeepSeek's documentation

What This Means for Developers

1. Batch your heavy jobs for off-peak hours. If you are in the Americas or Europe, this is largely automatic. For Asian developers, schedule long-running batch jobs outside 9–12 and 14–18 Beijing Time.

2. DeepSeek V4 remains the price leader. Even with peak doubling, it is 3–15x cheaper than GPT-5.6 or Claude Sonnet 5 for equivalent token volumes.

3. Peak pricing could become industry standard. If DeepSeek's model succeeds at managing compute demand, expect other providers — especially those with GPU constraints — to adopt similar time-based metering.

4. The 1M context window is now table stakes. With Gemini 3.1 at 2M, Qwen 3.7 at 1M, and now DeepSeek V4 at 1M, any new frontier model launching without at least 1M tokens of context will look outdated.

Bottom Line

DeepSeek V4 official is not a new model — it is a graduation. The preview has been available since spring, and the improvements are incremental rather than revolutionary. But the peak pricing model is genuinely new for the LLM API market, and it signals that compute supply is becoming the binding constraint even for Chinese labs with favorable GPU access.

For teams already on DeepSeek, the migration is low-friction. For teams evaluating providers, V4 official with 1M context at these prices keeps DeepSeek as the clear value play for high-volume workloads.

Comparing LLM pricing and capabilities? Browse our AI Model Directory for detailed specs, benchmarks, and pricing across 5,800+ models.