OpenHuman just smoked Hermes Nous Research 💀 same task. same prompt. straight speed test 😉

@NousResearch

@NousResearch lmfao, Poopjeet tries to shade hermes and forgets to blur his name and the model

@NousResearch This is deceptive. On the right you have Deepseek, one of the fastest local models available vs GLM which runs at half the tok/s on the same hardware. Do better.

don't know why the hermes cabal gets so agitated. 🤷‍♂️ here's the same model. openhuman is built on rust, uses a smaller system prompt, uses Jev (@typesafeai) and has a pretty smart token compression to keep costs low

@NousResearch So dishonest, different model. Crazy how a flash model is the faster one!

@NousResearch A very disingenuous post, using two separate models. Even if this was an honest test, I would take quality of work over speed, which Hermes consistently has given me ¯\_(ツ)_/¯

@NousResearch Here’s your award

@NousResearch Ain't that a debut and farewell at the same time?

@NousResearch This doesn't make you look good

@NousResearch There is still time to delete this

@NousResearch At least use the same model for both tests and run both agents at the same time: Hermes Agent is using GLM 5.3 Flash (medium), while Openhuman is using DeepSeek v4 Flash 0731 without specifying the reasoning level.

@NousResearch You might actually be retarded

@CommunityNotes the post shows a different reasoning level as well as different model. Thats false advertising.

哈哈哈哈哈，GLM 5.3flash 打 deepseek v4.1 flash🤣🤣🤣，我目前认为最慢和最快的模型打

@NousResearch I mean, first off, beautiful animation. Maybe stick to that. You didn't even manage to get Hermes agent counting, let alone what everyone else said here.

@NousResearch Tell all of X that you're an underhanded, deceitful organization without telling us. Whichever idiot on your staff thought this would be a good idea to post should probably be promoted to customer.

@NousResearch Damn this is pretty embarrassing. Bottom of the barrel vibe coding brain in action here in this video.

@NousResearch Zealy has ended, when's reward distribution

@NousResearch bro ain't no way I'm using some degenerate pokemon as my harness

@NousResearch Speed test ⭐💪

@NousResearch hey @grok what's the average tok/s in many public tests for GLM 5.3 flash and Deepseek-V4 Flash? Is Deepseek often ~2x as fast in these tests? Based on that, is the difference in speed in the video in favor of Hermes Agent or OpenHuman?

*You forgot DIFFERENT MODEL DIFFERENT REASONING LEVEL Nice too see a new company try to scam users with manipulated data.👍

@NousResearch At least next time try to make it more realistic because those experiment failed, I put my Nissan vs Lamborghini and Nissan wins , Awwww Nissan was the GRT tune-up

@NousResearch Could have at least used AI to match the models in your video editing process...

@NousResearch Make the test with same model on both Anyway, speed in agentic task is not the goal Stunt failed ( badly )

@NousResearch Tiny humans has tiny brains, confirmed

@NousResearch And you call your self an AI Lab !?

@NousResearch Interesting quality discrepancy I wonder what information you were trying to obscure in the beginning…

@NousResearch I tested two car drivers, I gave a Lambo to one driver and the other driver had the newest Nisan Sentra. Guess which driver won?

@NousResearch

@NousResearch Still choose Hermes

Openslop

@NousResearch Hay que ser Muy mogolico para usar dos modelos distintos y querer hacer una comparacion de harness

@NousResearch Jesus Christ man….

@NousResearch Hermes Is not a reasoning model. You choose it