- Sources: primary, discussion
- Summary: Artificial Analysis published its own measurements for DeepSeek V4 Flash 0731. It places the max-reasoning configuration at an Artificial Analysis Intelligence Index of 49.93 and at 0.027 USD per Intelligence Index task, the lowest cost per task of the 13 models in that chart, against 1.86 for GPT-5.6 Sol and 2.34 for Claude Opus 5 in the same configuration.
- Why it matters: Cost per task is the figure a team pays against when choosing an inference provider, and these are Artificial Analysis's own measured datasets rather than vendor-reported numbers.
send feedback on this story