DeepSeek states the 75% discount on its V4-Pro model is now permanent, lowering API costs well below major Western rivals and potentially reducing operating costs for high-token AI and crypto-related agent applications.
DeepSeek said Saturday that the 75% discount on its flagship V4-Pro model, previously scheduled to expire on May 31, will now remain in place permanently. According to the company, V4-Pro API pricing will stay at one-quarter of launch rates, with costs ranging from 0.025 yuan to 6 yuan per million tokens depending on usage type, including $0.87 per million output tokens, down from $3.48 at launch a month earlier. The report compares that rate with higher output-token prices from Western models including Claude Opus 4.7 at $25, GPT-5.5 at $30, Gemini 2.5 Pro at $12, Claude Sonnet at $15, and GPT-4.1 at $8 per million tokens. The article says a workload of 100 million output tokens per month would cost about $87 on V4-Pro versus about $2,500 on Opus 4.7 and $3,000 on GPT-5.5. DeepSeek did not say whether the permanent cut was enabled by Huawei Ascend 950 chip supply, though it had previously linked future lower pricing to larger shipments of those supernodes in the second half of 2026.