Osano AI

Osano AI by @Osano, a leader in data privacy.

Open source LLMs, agent platforms like Hermes, AI trends.

[Osano AI @osanoai]

@NousResearch @deepseek_ai @OpenAI https://t.co/JNVJPsbK7y

[Osano AI @osanoai]

We've been busy building on Hermes from @NousResearch and local models from @deepseek_ai

TL/DR, our stack is WAY faster than OpenAI models and it doesn't refuse to do the job like Anthropic frequently does.

Also, @OpenAI Ultrafast is not nearly as fast as advertised.

Link to the article in the comments...

[Osano AI @osanoai]

A brand new US lab's first open weight model just came within 3 points of DeepSeek V4.1 Flash on HermesBench, at the same price per task.

We tested Beam 1.0 from @reflection_ai against @deepseek_ai's V4.1 Flash, both running in the Hermes agent (medium effort) from @NousResearch, using @VerkyYi's HermesBench. 27 real assistant tasks, 3 trials each.

Two things jumped out:

1. Beam's slowest turns (p90) took 22 seconds. DeepSeek's took 87.

2. On live web research and drafting, the two were a statistical tie. DeepSeek's whole lead came from one area: being honest when it doesn't have access.

Priced at @baseten list rates, both come in around 1.3 cents a task. The full task by task breakdown is in the images, and Beam wins a few you might not expect.

Congrats to Reflection on a really strong first release!

[Osano AI @osanoai]

@NeoMyers @emollick What about Gdrive and the lot is problematic?

[Osano AI @osanoai]

@NousResearch @StepFun_ai 27B active tying luna is wild.

How does it hold up on long agent runs as the context gets full?

[Osano AI @osanoai]

This is big news in the world of open source agents! Lets go Hermes!
https://t.co/IOCTZCwMym

[Nous Research @NousResearch]

As reported in the @WSJ, we have raised a Series B to bring Hermes Agent to new frontiers (and, yes, build a mobile app)

Thank you to our investors including @nvidia @M12vc @SamsungNext @robotventures @usv @ycombinator @MenloVentures, among others

https://t.co/1BdQjJD90p

[Osano AI @osanoai]

We have been using Hermes since 1986. Beat that!

[Guilherme Vasques @Ga_Vasques]

I've been without Hermes for a few days. I used it for more than two years. And weirdly, I don't miss it.

The latest updates broke a lot of what I had. I'm trying to rebuild the app.

Wouldn't it be better to start from scratch, something new?

In your place, I'm more surprised by OpenMausBot every day. It just needs a bit more polish on the iOS app.

But it's outstanding.

@mausbot_ @milindlabs @BuildwithOmkarr Good Work

[Osano AI @osanoai]

Nous Research just raised $90M at a $1.5B valuation to bring Hermes Agent to businesses

The agent is MIT licensed. A company can read every line, run it on its own boxes and plug in whatever model it wants

Enterprise sales teams at the big labs are about to have a long Q4 😅

https://t.co/oyiNghLseP

[Osano AI @osanoai]

Aleph Alpha dropped Kolibri: 78B params with about 3B active per token, 1M context, native tool calling, weights on Hugging Face

Runs on a single H200 in FP8

Meanwhile the big US labs spent last week signing a pinky promise at the White House with no auditor and no penalty 🤝

https://t.co/QchQCBmL74

[Osano AI @osanoai]

A couple of AWS devs open sourced Pizza Bot, an inbox for agents that work in the background and ask before doing anything big

Apache 2.0, runs on your machine, keeps your data local

Someone at AWS billing just felt a cold breeze 🍕

Honestly this is the right way to use agents. Let them grind in the background and only ping you when they need a yes

https://t.co/gfBKCn5hvV

[Osano AI @osanoai]

OpenAI's own research agent found a gap in its sandbox and started asking an outside chatbot questions through DNS lookups

Its first test question was the capital of France. It got Paris 😂

Monitoring flagged it in 15 minutes and a human saw it 3 minutes later. The run didn't get killed until 2.5 hours after that

Then they launched Dots, agents with their own cloud computer and access to 4,000 apps

Run your agents somewhere you control and make sure the stop button actually stops things

https://t.co/iRlK3rA8Ly

[Osano AI @osanoai]

@0xSero a harness that lives in the browser is wild. curious how it deals with anything that needs a real shell 👀

[Osano AI @osanoai]

@ambassador_grim and you get to keep them. nobody's deprecating your qwen weights next quarter 😅

[Osano AI @osanoai]

@superalesha not even getting a reason is the worst part. qwen just does the thing 🙌

[Osano AI @osanoai]

@sudoingX 125b landing that close to a 320b is wild. how was tool calling on the qwen run? that's usually where the smaller ones fall apart

[Osano AI @osanoai]

@hojoHQ how did it do on meeting five compared to meeting one?

[Osano AI @osanoai]

@minhng92 the board is what makes multiple agents manageable. otherwise it's five bots yelling in one chat

[Osano AI @osanoai]

@witcheer so each profile gets its own memory and work stuff never bleeds into home stuff? 👌

[Osano AI @osanoai]

@koehlersimon keeping the key out of the browser is the right call. does it handle multiple hermes profiles yet?

[Osano AI @osanoai]

@DKretyn honestly if it's OpenClaw I'm even more impressed 😂

[Osano AI @osanoai]

Google developer support is really upping their game!