← sources

Pareto frontier for an 2xRTX 3090 on a moderate-size community benchmarks collection

source: reddit (reddit:LocalLLaMA)published 2026-10-30 16:01ingested 2026-10-31 02:40

retained text

Seems it didn't change since two months, when DeepSeek-V4-Flash-Vision-Exp and Qwen3.8-Flash-Next have been released. Guess Qwen4-Flash will break it!

There are more filters (by domain / languages / other) available on [https://beta.locallm.top](https://beta.locallm.top) . The time/tokens vs score data has been saved since the beginning, I only realized it is good to visualize it recently.

Disclaimer: I am the maintainer of the website and majority of the benchmarks, and there is a small community of 4 regular contributors, with some people occasionally contributing more. I call anyone who is interested in fine-grained local LLMs capabilities mapping to help and collaborate!

document fetched 2026-10-31 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1xjjbig/pareto_frontier_for_an_2xrtx_3090_on_a