Image: DeepSeek DeepSeek's V4-Pro Discount Is Now the Official Price
DeepSeek's 75% promotional discount on V4-Pro expired May 31, and the company made it permanent instead of reverting. Output tokens are now $0.87 per million.
DeepSeek ran a 75% discount on its V4-Pro API since the model launched, with a stated expiration of May 31, 2026 at 15:59 UTC. That deadline passed today, and the price did not go back up. The DeepSeek API pricing page now lists the discounted rates as the official prices, not promotional ones.
The new permanent rates for DeepSeek-V4-Pro:
| Before discount | Now permanent | |
|---|---|---|
| Input (cache miss) | $1.74/M tokens | $0.435/M tokens |
| Input (cache hit) | $0.0145/M tokens | $0.003625/M tokens |
| Output | $3.48/M tokens | $0.87/M tokens |
The competitive context
Output at $0.87 per million tokens puts V4-Pro well below most frontier alternatives. GPT-5.5 output runs around $30/M. Claude Opus output is similar in range. On raw token cost, V4-Pro is roughly 34x cheaper than GPT-5.5 for output.
The performance gap doesn’t justify a 34x price premium for a large class of tasks. On coding benchmarks, V4-Pro lands within a few percentage points of the top frontier models. For automated tasks where you’re routing thousands of requests, generating code, processing repositories, or running analysis pipelines, V4-Pro at permanent discount pricing changes the math considerably compared to six months ago.
Cache hits are especially significant. If you’re running a fixed system prompt or feeding the same large context repeatedly, the cache-hit price drops to $0.003625/M on input. A system that reuses a 6,000-token prompt with 90% cache efficiency sees effective input costs drop well below a dollar per million tokens.
What changed for DeepSeek to make this permanent
DeepSeek hasn’t published an explicit statement explaining the decision. The most straightforward interpretation: the discount was always intended to drive adoption, it worked, and the company decided its long-term price point was closer to the discounted rate than the launch price. This is the same playbook AWS and others used when cloud compute prices fell faster than anyone publicly expected.
There’s also competitive pressure. Several other high-performance, low-cost models have launched or reduced prices this year. Letting the discount expire and reverting to $3.48/M output would have created an obvious comparison point for anyone evaluating alternatives.
Practical implications for developers
If you’ve been evaluating DeepSeek-V4-Pro but waiting to see whether the discount was real, the question is settled. The rates are the rates.
The more interesting question is how this affects routing decisions for teams using multiple models. A growing number of AI coding tools and agent platforms let you configure which model handles which request type. At $0.87/M output with near-frontier coding performance, V4-Pro is a reasonable default for a lot of workload types that don’t specifically require GPT-5 or Opus 4.8 quality. The premium models still have ceiling advantages on complex reasoning and very large context tasks, but a lot of production code generation doesn’t hit that ceiling.
DeepSeek also cut input cache-hit prices across its full model lineup to one-tenth of launch pricing in April. The permanent V4-Pro cut follows that direction.
Sources: DeepSeek API Pricing, APIdog — DeepSeek V4-Pro permanent price cut, Startup Fortune