Charting Model Intelligence vs Cost
krishnan · x · 2026-07-20
This post includes a "Intelligence vs Cost per Task" scatter plot from Artificial Analysis. It maps 576 models based on an intelligence index (y-axis) and cost per task on a log scale (x-axis), illustrating how different vendors balance "higher capability" with "lower cost".
The chart highlights Kimi K3, placing it in the high-intelligence, lower-cost region, suggesting it is highly competitive in such comparisons. Ultimately, the information points to a core question: why do US frontier AI labs invest so much money, and how do these investments reflect on model capabilities and inference costs.
More from Infra
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27
- Open-source Latch compiles PyTorch into training megakernels, 2.75x faster on H100 — sguada · 2026-10-27
- NVIDIA claims HGX Rubin NVL8 delivers 10x tokens per megawatt vs B200 — thursdai_pod · 2026-10-27
- NVIDIA Claims HGX Rubin NVL8 Delivers 10x More Tokens per Megawatt Than B200 — thursdai_pod · 2026-10-27
- Together lets multiple LoRA post-training experiments share the same GPU capacity — togethercompute · 2026-10-27
- Commoditization dilemma: either human labor or GPUs become the commodity — _AustinCalvert_ · 2026-10-27