Hello everyone! I'm a preset creator for AIRP; and after I made a reddit post on my many community subreddits asking for people to please vote when they can; I was directed to also post here! (So I am sorry if this is wildly inappropriate) but

We need your votes!

I hired a professional with a masters degree in AI/ML (pursuing their PhD) to help make a benchmark for us, for RP. Now as we all know; a benchmark for RP will never be perfect because everyone RP's differently, like different things, yada yada yada. That's why we tried to focus on a few objective landmarks, as well as a an Arena style versus bench.

First up: The LLM Arena. We show you two different rewritten responses from two different models who were given a chat and asked 'please take this turn.' We then show that to you. Does it have a preset attached? No; because that's another variable that'd be added and for our own safety we aren't benchmarking presets. (LMAO. We'd be dead in the fucking streets by sundown.) You judge of the two responses which you like more. Thaaaat's it folks. We literally cannot game the results as it is people just blindvoting which they like more. (So if you see something you don't like IT'S NOT UP TO US.)

If you have ever used 'LMarena' it's the same thing; but for RP!

On the other end we also score an LLM in a similar vein on how they follow certain instructions and if they maintain consistency. Since 'did it take users agency' is an objective yes or no answer; we handle these tests with an LLM judge and the benchmark overseer (Levi is his name) monitoring the arbiters answers. (In our case; Sonnet.)

This bench is/was expensive. It cost money to run these models through this gauntlet; and we don't get much out of it. We publish all the data for everyone to see, and all we want to do is help our community out. If you have any questions on how we grade things; head to our methodology page! You'll notice somethings are a bit LLM-Written. This is not because Levi isn't a professional; but rather because he's ESL; so when publishing something important like this he wanted to make sure all his ideas and such were properly translated. So be nice!

You don't have to log in, you don't have to do anything. Just read a chat and vote on a thing. I know you've got an opinion; so share it. Big companies ignore us RPers all the time. Is a bench for RP as useful or objective as one for coding, web-dev, or math? No. But that doesn't mean we don't deserve one, or that it has no value at all. So please; help a bunny out. Drop a vote! The data is only as useful as you help make it. We have 800 votes and we want to get about 3,000 this time around; so we can start our next benchmark. (Testing models across their 'lineages'. So Opus 4.6 vs 4.7 vs 4.8; Deepseek 3.2 vs Deepseek v4 Pro, fun stuff!)

But we can't do that till this one closes! And if it can't hit enough votes; we'll know that this kinda stuff just isn't what the RP community wants. The only thing that'll ever make Plotpoints go down is a lack of community engagement/inability to get new votes.

Don't see a model you expected to? That's cause it's either new, or we didn't have the money at the time to run it. If the community likes this; we will add more. (And take requests! Please only suggest models available on OpenRouter though; we source all our models from the same platform to reduce variables. I know OR has a lot of local ones!)

All we need you to do is head to this website; and vote between pairs.

Huggingface Link: https://huggingface.co/spaces/RoleCall/rolecall-studios (Go here to make model requests!)
Vote Link: https://plotlightstudios.com/plotpoints/multiturn
Result Hub Link: https://plotlightstudios.com/plotpoints
Methodology Link: https://plotlightstudios.com/plotpoints/methodology
Github Link: https://github.com/LeviTheWeasel/rp-benchmark
Discord Link for all our sites: https://discord.gg/aFvkTCDRtf

Where can you make a model requests (With OR Links please!!)