DeepSeek has stunned the market with a 12x price surge for its flagship V4-Pro model, effective August 16. Peak-hour input tokens will leap from $0.003625 to $0.044 per 1M—a shockwave mitigated only for yuan-paying users, as the exchange rate masks the hike’s severity. Even output tokens, rising 4.5x in both currencies, feel less punishing locally. Off-peak rates, though still up 6x, offer slight relief.
The overhaul introduces time-based discounts: off-peak hours (outside 04:00–07:00 & 09:00–13:00 MSK) slash costs by up to 50%. V4-Flash input tokens now cost $0.007 off-peak (up from $0.0028) and $0.014 during peaks—a 2.5x to 5x jump. Yet DeepSeek remains cheaper than Google’s Gemini 3.7 Flash ($0.075 per 1M input tokens), despite the latter’s edge in TerminalBench scores (87.9 vs. 85.8).
Enter Muse Spark 1.2, the wild card. Its contributor tier undercuts DeepSeek’s new off-peak Flash by 3.5x ($0.002 vs. $0.007 per 1M input tokens), with competitive benchmarks (DeepSWE: 59.3 vs. Flash’s 54.4). The caveat? Regional API limitations and a steep non-contributor rate ($0.15 per 1M tokens). For cost-conscious developers, DeepSeek’s Pro still beats Gemini on price, but Muse Spark’s budget option could be the game-changer—if accessible.