Hermes Agent Launches Mixture Of Agents 2.0 For Custom Model Combinations

We are working on benching various combos of open source models to see if we can get Opus levels with much cheaper models as well, stay tuned!

We have Fable at home

@Teknium Please advanced people of Hermes, how can a normal user have an Hermes assistant spending less of 30$/month? Please somebody speaking about providers and subscriptions? Opencode-go is short for me, Nous Portal too expensive, I didn't try Codex oauth, best compatible providers?

@Teknium ok this is fucking huge

holy fuck epic work @Teknium

@mossy98_ No - we had an MoA tool in Hermes since launch - which was an idea by TogetherAI 2 years ago, we just revamped our MoA tool into virtual models, made them agentically orchestrateable, and benched

https://arxiv.org/abs/2406.04692

@eiomsirius @Teknium Try Neuralwatt. And Oracle Cloud for hosting. I spend less than $20 a month on GLM 5.2. I’m actively waiting for Neuralwatt to be a native provider with energy usage tracking.

@Teknium Can this be done with local models? That would be sick.

@Teknium Can we mix open source models? Or only frontier?

@FabiusDefi "in our soon-to-release HermesBench against opus and gpt-5.5 with MoA using Opus & GPT together."

its coming bro lol

@Teknium We need see how low-cost models combo perform (deepseekV4 flash)

everyone's stuck waiting on gpt-5.6 access today. meanwhile hermes agent just exposed its MoA presets as virtual models, nous says 8% over opus 4.8 and 11% over gpt-5.5 on their upcoming benchmark.

@Teknium Is this like Open Router Fusion?

@eiomsirius Our hope is that mixture of agents will allow mixing cheap models to get the performance of strong ones

@Teknium I get the point! Than you, good job.

I would also ask the people making content if they can bring some light about best providers right now, compatibilities, cheap subscription plans for "normal people", since I think Hermes is going much beyond developers and tech people. 👁️

@Teknium 8% higher than opus through moa is impressive but does the latency cost of running multiple models actually make this practical

@hector_podcast Any model any amount of models and any provider! We just had solid final results for this combo today, early next week we should have a much wider distribution of mixtures to share benchmarks for

@Teknium @GoDjMike Nice! This of the next wave, mixture of agents

@Teknium bench numbers or vibes though?

@Teknium drop the bench