Raju Datla

@Teknium Yep, Hermes main brain with Gpt 5.5 is magic, but then most anything with Hermes, other than some open chineese models works fairly, just not as great at GPT 5.5 though at this point

In case there’s any confusion, Hermes has always had the ability to use any sub you have with OpenAI, and there is no distinction between codex and chatgpt subscriptions. Try it today with Hermes and see for yourself 🤗

@imukulmunjal True, this person, didn't even remember what version of Codex she used last and I don't knows 5.5 is out. And yes I asked the thinking process and then I didn';t know how to end it quickly, since I knew she was completely clueless. Funny thing is she organizes AI meets

@raju_datla1 screening for AI agentic roles is genuinely broken right now. resume tells you nothing. you need to put them in an unscripted problem and see how they think, not what frameworks they list

Ok, just did an interview to hire AI AGentic engineer or how I now define product engineer. A general Java, mid level engineer and it was very frustrating experience. Holy cow, most software engineers no wonder are lost in AI models doing most of the coding. The problem is don't even know how to screen resumes these days!

@mattshumer_ Just click baits or attention w.... types, who just want more attention. I try to ignore but hard to sometimes.

Totally false. Who still thinks this?

@eliana_jordan Yeah, too late to feel the pain. Go Hermes, if you want much less painful agentic harness experience. OpenClaw is pretty good and easy to use, but very painful for consistency or high level work or enterprise level work or serious business use case in other words.

installing openclaw today finally am i too late?

@JohnShina91 @quxiaoyin DeepSeek to debug Opus 4.7, okie dokie then, its barely even out for 2 days or so?

@raju_datla1 @quxiaoyin What I use DeepSeek for is to debug codes written by the top dogs - Opus 4.7, GLM 5.1, K2.6. None of these debug without causing regression. Knowing what each does best would save us from making false claims about models that we've not tested.

I can’t believe I stopped using Claude Code max and entirely use DeepSeek and Hermes. It’s so fast, so so fast, 3x faster for the same task. So cheap. I spent $5 last week and never need worry about being rate limited or usage hit limits very two hours. For most tasks it’s perfect enough.

@pedro7k2 @quxiaoyin Yeah, don't assume things, without knowing what I do.

@raju_datla1 @quxiaoyin If you're working on heavy-duty coding tasks, Chinese models might not cut it. But the fact that you listed MiniMax as a viable option makes me think you haven't actually tried them yet.

@ChinesePowered @quxiaoyin Yeah, thats my experience, GLM5.1 is great for even orchestration etc, since I use it extensively in Hermes. That and Minimax seems to be pretty good with agentic work

@quxiaoyin @raju_datla1 As a fellow Chinese, I recommend GLM for coding. Qwen for multi modal.

Because Most indians look for paycheck, especially in India, not chase something great. I am about help another company get rid of another indian outsourcing company, because their quality is piss poor and don't want to learn AI Agentic frameworks. Its a different story of Indians in USA.

why does india not have a single serious ai model for coding or general use, every major drop is still from usa or china what are we doing with soo many devs

@farzyness Well thats the issue with many Sr Software engineers, still stuck in the ego of they are better than AI Models still. So no fear, plenty of company there with what Karpathy is saying :-)

This hits my soul so hard.

@brnnnmcdnld @quxiaoyin 5.5 is not even comparable to any other model, so I wont expect that. If its its close to Opus 4.7 or even 4.6 then I am impressed, not that I like Opus models anymore, other than using them for review for now.

@raju_datla1 @quxiaoyin Honestly I was surprised, given the cost difference alone the outstanding gaps between say, V4 Pro and ChatGPT 5.5 are forgivable.

So this one works based on say if I have like a sprint or stories, just use /goals sprint/story xxxx and then it self checks and verifies the story/sprint for completion including like QA/testing? Or more of you have to create specific requirement to fit this, kind of like Ralph loop?

/goal also lands in Codex CLI 0.128.0. Our take on the Ralph loop: keep a goal alive across turns. Don't stop until it's achieved. Built by my co-worker and OpenAI mentor Eric Traut, aka the Pyright guy. One of the GOATs I get to work with daily.

@brnnnmcdnld @quxiaoyin Ok,well I guess I will need to evaluate again and see if it at least caught up to GLM5.1. I have now seen multiple post this type of thing with Deepseek so maybe need to see if I stand corrected.

@raju_datla1 @quxiaoyin I've been testing out DeepSeek V4 Pro this week for code review tasks and other refactoring type evals I have. It's really not that far behind SOTA. I don't think you can fairly make a claim that it's not competitive unless for non-coding work or "non-serious" work...

@JeekbeekVin @rohanpaul_ai Holy cow the ignorance..... Dripping... And ego also. Ok, live in that world, questioning arguably the guy pretty much launched the current AI revolution or at least one of them.

@raju_datla1 @rohanpaul_ai What has he actually built software wise with this human out of the loop system he is speaking about? I saw him drop auto research and a obsidian based knowledge system . What else has he produced that indicates he is using AI to build complicated software systems?

Andrej Karpathy: "To get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You cannot be there to prompt the next thing. You need to take yourself outside the loop. You have to arrange things such that they are completely autonomous. The more you can maximize your token throughput and not be in the loop, the better. This is the goal. So, I kind of mentioned that the name of the game now is to increase your leverage. I put in very few tokens just once in a while, and a huge amount of stuff happens on my behalf." --- From @NoPriorsPod YT channel (link in comment)

@rezoundous Yep, only choice with Hermes main agent/orchestrator is Gpt.5-5, used to be Opus was the best one, but 5.5 just hands down is now the best one to use.

If your Hermes agent is not using GPT-5.5, I'd recommend you try it out.

@0Sunnycity2 @garrytan Just do it. Overthinking in AI world is not useful or productive. It can always be removed if it doesn't work. So don't get stuck in past on how difficult it might be remove etc.

@raju_datla1 @garrytan Thinking of setting it up. How is the experience with Hermes?

GBrain v0.23.0 is actually... a quite big deal. Your brain actaully gets enriched from your conversations with OpenClaw now. OpenClaw still has its own separate memory, but the brain's people, companies, and concepts now all benefit from what it learns from you.

@iScienceLuvr Hmmm.. Use Hermes or Openclaw for remote. I mean its not that complicated. YOu don't need Claude/Codex to provide that anymore

Claude Code has remote control which is amazing but has low rate limits. Codex has high rate limits but no remote control. How do I get the best of both worlds???