Who is better between DeepSeek V4 Pro and Grok 4.6? @grok In the latest results of Artificial Analysis, the Intelligence Index of Grok 4.6 High is 61, and DeepSeek V4 Pro 0813 Max is 53. Looking solely at overall intelligence, Grok 4.6 is clearly ahead and already in the same league as GPT-5.6 Sol Max. Grok's self announced GDPVal AA v2 reached 1753, even surpassing GPT-5.6 Sol's 1728 and Fable 5's 1741, demonstrating strong knowledge work, complex reasoning, and synthesis tasks. But when it comes to agents and real work scenarios, DeepSeek begins to become terrifying. On DeepSWE, Grok 4.6 is 65.9% and DeepSeek V4 Pro is 62.7%, which is very close. DeepSeek achieved 87.9 on Terminal Bench 2.1, while Fable 5 achieved 88.0, CyberGym 83.3，Toolathlon-Verified 74.1。 This generation of V4 Pro is clearly not simply pursuing smarter exams, but is crazily supplementing tool calls, terminal operations, code agents, and long task execution capabilities. Then looking at the price, the gap suddenly widened, and my chin was shocked The DeepSeek V4 Pro API has an input of $0.435 per million tokens, an output of $0.87, and a context of 1 million. Grok 4.6 has an input of $2, an output of $6, and a context of 500000. Artificial Analysis ran the complete Intelligence Index, with Grok 4.6 costing $1068.47 and DeepSeek V4 Pro costing only $135.03, the former being close to 8 times the latter. Even more outrageous is that DeepSeek's measured output speed is about 83 tokens/s, while Grok's is about 66 tokens/s. So this time, I actually feel that whether DeepSeek or Grok is stronger is no longer the most valuable question. If you are pursuing the strongest comprehensive abilities, knowledge work, multimodality, and ability limits currently available, Grok 4.6 is even stronger. But if you plan to have hundreds or thousands of agents write code, run terminals, debug tools, and handle repetitive tasks every day, the economics of DeepSeek V4 Pro are too exaggerated.