DeepSeek V4 Pro pricing rules
Cache-hit and cache-miss input rates are separate; supports thinking and non-thinking modes.
Official source| Fact | DeepSeek V4 Pro | GPT-5.6 Terra |
|---|---|---|
| Input / MTok | $0.4350 | $2.50 |
| Output / MTok | $0.8700 | $15.00 |
| Cached input | $0.003625 | Not recorded |
| Context | 1,000,000 | 1,050,000 |
| Example request | $0.00609 | $0.0550 |
| Verified | 2026-07-16 | 2026-07-16 |
Cache-hit and cache-miss input rates are separate; supports thinking and non-thinking modes.
Official sourceOfficial list rate; undocumented cache and special tiers are omitted from calculations.
Official sourceDeepSeek V4 Pro has the lower list-price result for 10,000 input and 2,000 output tokens. This is not a quality ranking.
It can. Context limits determine eligibility, and documented long-context thresholds can apply higher rates.
The baseline example excludes cache so the standard rates are comparable. Use the calculator to enter an expected cache hit.
Only when the provider documents batch eligibility and the workload can tolerate asynchronous processing.
No. Search, code execution and other server-side tools can carry separate provider charges.
Price alone cannot determine best. Evaluate capability, latency, reliability, data policy, availability and task-specific performance separately.