DeepSeek V4 Pro keeps the 75% API price cut as a permanent agent-cost lever
DeepSeek's pricing page now says V4 Pro will keep the post-promotion 75% discount as the official price, creating a practical cost-control option for high-token agent workflows.
DeepSeek V4 Pro pricing in plain terms
DeepSeek's official API pricing page says the DeepSeek V4 Pro API price will be adjusted to one quarter of its original price after the current 75% discount promotion ends on May 31, 2026. The listed discounted prices are $0.003625 per 1M input tokens on cache hits, $0.435 per 1M input tokens on cache misses, and $0.87 per 1M output tokens. OpenRouter separately lists DeepSeek V4 Pro with matching prompt, completion, and cache-read pricing, which makes the change relevant for teams comparing direct API access with router-based access.
Key takeaways
- DeepSeek's docs list V4 Pro with a 1M-token context window, thinking and non-thinking modes, tool calls, JSON output, and a maximum 384K output length.
- The official pricing note says the V4 Pro API price will become one quarter of the original price after the 75% discount promotion ends on 2026-05-31 15:59 UTC.
- OpenRouter's model page corroborates the V4 Pro price points at $0.435 per 1M prompt tokens, $0.87 per 1M completion tokens, and $0.003625 per 1M cache-read tokens.
- For coding agents, the most important detail is not only the headline discount; it is the low cache-hit price when repeated context can be reused.
- The Hacker News thread shows developer interest, but the actionable numbers should be verified against live pricing pages before purchase or migration.
Practical LinkLoot angle
This is a deal story only if your workflow burns a lot of tokens. A permanent V4 Pro price cut can matter for agent loops that repeatedly send repository context, long logs, retrieval bundles, or tool transcripts, but it does not automatically make DeepSeek the best choice for every task. The useful play is to route expensive context-heavy steps to the cheapest model that still passes your quality gate, while keeping frontier models for tasks where failure costs more than inference.
| Option | Best use | Limitation to check | Source |
|---|---|---|---|
| DeepSeek V4 Pro direct API | Long-context agent work where cache hits and low output cost matter | Direct billing, rate limits, and compatibility with your SDK stack | DeepSeek API Docs |
| DeepSeek V4 Pro via OpenRouter | Comparing providers, centralizing model routing, or testing without changing app billing paths | Router availability, provider routing, and any extra platform policies | OpenRouter |
| Existing frontier coding model | High-risk refactors, security-sensitive work, or tasks where your team already has reliable evals | Higher cost may still be justified if it prevents failed agent runs | Internal benchmark needed |
A practical workflow is to benchmark three task types separately: codebase search and planning, implementation with tests, and final review. V4 Pro's long context and cache economics may shine in the planning stage, while another model may still be better for the final patch or security review.
What to verify before you act
Check the live DeepSeek pricing page before changing budgets, because the same page says product prices may vary and recommends regular checks. Confirm whether your workload can actually benefit from cache-hit pricing; if every request sends fresh context, the cache-hit number is mostly irrelevant. Also verify output quality with your own repository or workflow transcripts, because low cost can disappear quickly if the agent needs repeated retries.
If you use OpenRouter, compare the model page with the direct DeepSeek page and run a small billing test before routing production traffic. The listed OpenRouter prices corroborate the direct V4 Pro figures, but platform routing, availability, and provider-specific behavior can still affect latency, limits, and reliability.
DeepSeek's pricing page says V4 Pro will keep the 75% discount level as the official price after the promotion ends on May 31, 2026.
For more ways to evaluate cost-effective automation stacks, see LinkLoot's guide to lifetime software deals and use the same discipline for API pricing: durable usefulness first, discount second.
Source check
DeepSeek's official docs confirm the V4 Pro feature table, context length, listed token prices, and the note that the 75% discount becomes the post-promotion official price. OpenRouter independently lists DeepSeek V4 Pro and corroborates the same per-token pricing structure for prompt, completion, and cache-read tokens. Hacker News confirms that developers are actively discussing the pricing change, but it is used only as a signal of attention, not as proof of the billing terms.