Nous Research #announcements

@everyone

Nous Portal is experiencing degredation and outages at the moment, will update when we learn more or issue is resolved.

Nous Research #announcements

@everyone

Services are restored!

Superposition #announcements

@everyone superposition chain wind down has now commenced! if you missed the boat to bridge assets out, please create a ticket ASAP. more announcements coming soon. thank you everyone

Nous Research #hermes-announcements

@hermes-agent-notifications

GPT-Image-2.5 now available in Hermes Agent through Codex subscriptions and Fal AI!

Soon on Nous Portal!


https://fxtwitter.com/Teknium/status/2097465800231883091

GPT\-Image\-2\.5 now available in Hermes Agent through Codex subscriptions and @fal - https://x.com/fal\!
︀︀
︀︀Soon on Nous Portal\!

> Quoting - https://x.com/OpenAI/status/2097394956457623964 OpenAI \(@OpenAI - https://x.com/OpenAI\)
> ︀
> ChatGPT Images 2\.5—faster, sharper, smarter, with better tools for creating whatever you can dream of\.
> ︀︀
> ︀︀\- Faster image generation to keep your ideas flowing
> ︀︀
> ︀︀\- Improved fidelity for more natural, recognizable images
> ︀︀
> ︀︀\- Consistent details across multiple edits
> ︀︀
> ︀︀\- Comment\-based edits to change only what you want

💬 - https://x.com/intent/tweet?inreplyto=2097465800231883091 2 🔁 - https://x.com/intent/retweet?tweetid=2097465800231883091 1 ❤️ - https://x.com/intent/like?tweetid=2097465800231883091 5 👁️ 826 

Gensyn #︱📢︱announcements

New Research X Space

Join Oghuzan Ersoy at <t:1789056000> to talk through IR3DE-AXL - https://www.gensyn.ai/research/ir3de-axl-collective-inference-network, IR3DE - https://www.gensyn.ai/research/look-beyond-one-size-fits-all-llms-with-ir3de and AXL - https://www.gensyn.ai/news/introducing-axl combined to form a collective inference network!

Building on the earlier work, IR3DE-AXL allows participants to contribute expert models, use models served by others, or do both. A participant with a coding model can serve coding requests, while using someone else’s math, multi-lingual, or general-chat model.

Set your reminders here 👇

https://discord.com/events/852932483691577395/1547274717549830275

@Tech and Dev Updates

Gensyn | IR3DE-AXL: Collective Inference Network

Most inference routing runs through a single gateway with a fixed menu of models. IR3DE-AXL flips that: every participant builds their own router from whatever experts the network currently offers, and can contribute their own in return.

Nous Research #hermes-announcements

@hermes-agent-notifications

You can now install plugins from private github repos - Whether it's your orgs internal plugins or your own private ones, access them all easily in Hermes 🙂

https://x.com/Teknium/status/2097958504904736893

You can now install plugins in Hermes Agent from private repos\!

It'll use your stored GitHub credentials now\.

https://twitter.com/Teknium/status/2097958504904736893

Nous Research #hermes-announcements

@hermes-agent-notifications

Hermes Agent now displays detailed information on all subagent activities live, and you can steer and stop them manually from the CLI and Desktop Application.

https://fxtwitter.com/NousResearch/status/2098071687145365632

Hermes Agent now displays detailed information on all subagent activities live, and you can steer and stop them manually from the CLI and Desktop Application\.

👁️ 1\.4K 

Nous Research #hermes-announcements

@hermes-agent-notifications

Use DeepSeek Flash V4.1 in Hermes Agent through Nous Portal, OpenRouter, and DeepSeek Direct now!

https://x.com/Teknium/status/2098088984383725725

Deepseek Flash V4\.1 is now live in Hermes Agent through Nous Portal and others\!

https://twitter.com/Teknium/status/2098088984383725725

Nous Research #hermes-announcements

@hermes-agent-notifications

We just hit 3000 contributors. Everyone thank a @Developer today

https://x.com/Teknium/status/2098800012549619996

Hermes Agent has just hit 3000 contributors\.

Thank you to all of the developers who have worked to make hermes better for everyone\!

https://twitter.com/Teknium/status/2098800012549619996

Ambient #📢│announcements

@⁣       ↑ Notification Roles ↑        ⁣
🚀 WEEK 26 — MODEL SHOWDOWN
🎯 Theme: Which model should handle the job?
This week, we’re putting different models head-to-head.

The goal isn’t simply to find a “best” model. We want to understand which model performs best for which workload — and why.

🆕 What’s New
This week we’re introducing:

GLM 5.2 reliability improvements
Continued DeepSeek verification testing
GLM 5.3 Flash evaluation
Evaluation of newer Qwen models
Expanded community benchmarking


👤 USER LOOP — “Same Prompt, Different Model”
Take one prompt and run it across the models available to you.

Try testing:

🧠 Factual questions
💻 Coding
➗ Mathematical reasoning
📋 Instruction following
📚 Long-context tasks
📦 Structured outputs
🎨 Creative tasks
Then compare the results across:

Accuracy
Latency
Reasoning quality
Instruction following
Formatting
Consistency
Don’t just tell us which model you prefer. Tell us why.

Share your findings in https://discord.com/channels/1334942930695225365/1448970141940322354

🛠️ DEV LOOP — “Build the Benchmark”
Create a small, repeatable benchmark.

Keep the following consistent:

Prompts
Expected outputs
Scoring criteria
Test conditions
Then compare models based on:

Correctness
Speed
Reliability
Consistency
Failure rate
🔥 Bonus Challenge
Run the same benchmark during quiet and busy periods.

Do the results change?

If they do, that could reveal interesting differences in network conditions, routing, or infrastructure performance. https://discord.com/channels/1334942930695225365/1430214815833653361

⚙️ INFRA LOOP — “Model Health”
This week, pay close attention to:

GLM 5.2 stability
External miner reliability
Model-specific errors
Routing behaviour
Capacity differences between models
When something fails, try to determine whether the failure is related to:

A specific model
A specific workload
Network demand
External infrastructure
Request length
The goal is to move beyond “it failed” and figure out why it failed.

🌐 ECOSYSTEM LOOP — “Community Benchmark”
Have a challenge that other testers can reproduce?

Share it with the community. https://discord.com/channels/1334942930695225365/1430214815833653361

Include:

📝 Your prompt
🤖 Model tested
🎯 Expected behaviour
📊 Actual result
⏱️ Response time
🔁 Whether the result was repeatable
Other community members can then run the same challenge and compare their results.

Over time, these shared tests will build a community-generated picture of how each model performs in real-world conditions.

💡 WHY THIS MATTERS

Supporting more models only matters if we understand how those models actually behave in the network.

Different models have different:

Strengths
Latency profiles
Reliability characteristics
Infrastructure requirements
Community benchmarking helps us identify those differences while giving the engineering team real workloads and real failure cases to investigate.

More importantly, this brings Ambient closer to infrastructure that can make intelligent decisions about where inference should run.

🏁 This week’s mission
Test. Compare. Benchmark. Share.

Don’t just find the model you like.

Find out which model is right for the job — and prove it.

Nous Research #hermes-announcements

@hermes-agent-notifications

Nous Portal now offers ChatGPT-Image-2.5 for your Hermz

https://x.com/Teknium/status/2099327677648015365

ChatGPT\-Image\-2\.5 is now available on Nous Portal to use in Hermes Agent, as well as it's previously supported Fal and ChatGPT direct routes\!

https://twitter.com/Teknium/status/2099327677648015365

Ambient #📢│announcements

We have a public status page now: https://ambient.betteruptime.com

It checks GLM 5.2 (`ambient/large`) every 5 minutes with a real chat request, and only shows red if it fails from 3 of 4 locations for 5 minutes straight.

This is a pilot status page for the currently supported public model.

Thanks @jeff_okhihie for building the first community version of this idea 🙏

Ambient status

Welcome to Ambient status page for real-time and historical data on system performance.

https://ambient.betteruptime.com/

Nous Research #hermes-announcements

@hermes-agent-notifications

Hermes Agent is open for business.

Nous Portal now lets you invite colleagues to a Hermes Business account: your team gets agents across channels while sharing one central balance with per-member caps and shared skills that compound into proprietary IP.

Hermes Enterprise brings the same capabilities to on-prem or the cloud of your choice: a complete, self-improving, sovereign AI stack already trusted by some of the world's largest companies. Contact us to join them.

https://portal.nousresearch.com/business?utmsource=twitter&utmmedium=social&utmcampaign=businesslaunch&utmcontent=tweet2026-09-08

https://fxtwitter.com/NousResearch/status/2099599032037388404

Hermes Agent is open for business\.
︀︀
︀︀Nous Portal now lets you invite colleagues to a Hermes Business account\: your team gets agents across channels while sharing one central balance with per\-member caps and shared skills that compound into proprietary IP\.
︀︀
︀︀Hermes Enterprise brings the same capabilities to on\-prem or the cloud of your choice\: a complete, self\-improving, sovereign AI stack already trusted by some of the world's largest companies\. Contact us to join them\.
︀︀
︀︀portal.nousresearch.com/business?utmsource=twitter&utmmedium=social&utmcampaign=businesslaunch&utmcontent=tweet2026-09-08 - https://portal.nousresearch.com/business?utmsource=twitter&utmmedium=social&utmcampaign=businesslaunch&utmcontent=tweet2026-09-08

💬 - https://x.com/intent/tweet?inreplyto=2099599032037388404 44 🔁 - https://x.com/intent/retweet?tweetid=2099599032037388404 20 ❤️ - https://x.com/intent/like?tweetid=2099599032037388404 193 👁️ 9\.4K 

Gensyn #︱📢︱announcements

Introducing open-1b - The first language model with auditable, verifiable training.

open-1b represents a milestone on the path toward verifiable AI, a goal that is absolutely necessary for the future of intelligence.

The most used models are closed and concentrated amongst a few companies. How they were built, what went in and what did not is hidden and unknowable. Their biases are unknown and so unable to be trusted. The owners of those models suggest that they are the only that are responsible enough to be trusted with them.

Instead of having to trust how a model was trained, open-1b comes with its complete pretraining dataset, training and evaluation code, intermediate checkpoints at 100-step intervals, and a canonical state hash for every one of the 80,957 optimizer steps that produced it.

Anyone can load a checkpoint, replay the associated step on their own hardware, hash the result and confirm it with the published fingerprint.

AI verification is fast becoming a critical requirement. Models you can trust are the only defence against models that you cannot. Models that have their entire history on the public record and are able to be replayed.

Determinism means the same machine gives the same answer twice. Reproducibility means a different machine gives the same bits. Existing determinism settings make runs repeatable on the same hardware. That is not verifiable, as it is not reproducible by anyone else.

Our verifiable AI infrastructure allowed that gap to be closed. RepOps - https://www.gensyn.ai/research/verde-a-verification-system-for-machine-learning-over-untrusted-nodes, our library of reproducible operations, and REE - https://www.gensyn.ai/news/ree, our reproducible execution environment, make matrix multiply, normalization and gradient reduction produce identical bits whether it runs on a consumer NVIDIA card, an x86 or ARM CPU, or MacBook.

Auditing the training of open-1b is a collective exercise, and you can take part. Verifying all 80 957 steps alone isn’t practical, but with enough people reviewing enough of it, the whole is verified.

- Download the audit harness
- Pick any step of the run
- Replay it

It runs on NVIDIA GPUs, x86 and ARM CPUs, and natively on Apple Silicon. When your result matches the published hash, your verification is recorded and credited on-chain in the public ledger - https://open1b.gensyn.ai. This ledger assembles the individual checks into a single collective statement: this model was trained exactly as declared.

There is no reward or yield from auditing. The reward is your name on the immutable record of the first ever fully audited training run.

In a future that has achieved democratisation of intelligence, that is an important place in history.

7/

Read more here:

Blog Post: https://www.gensyn.ai/news/introducing-open-1b-auditable-training

Paper: https://open1b.gensyn.ai/open1b-tech-report.pdf

Github: https://github.com/gensyn-ai/open-transformers

Huggingface: https://huggingface.co/collections/Gensyn/open-1b

Auditing open-1b: https://open1b.gensyn.ai

Audit tool: https://github.com/gensyn-ai/pretraining-audit-cli

@everyone

Gensyn | Introducing open-1b: the first model you don’t have to t...

Auditable training is the best defense against the future of AI we’re being warned about. Gensyn has proved it’s possible.

Nous Research #hermes-announcements

@hermes-agent-notifications

New blog post:

We had a million lines of Python to clean up. On September 2nd I asked Hermes Agent to do it.

1,393 subagents and nineteen hours later, the codebase was 34.4% smaller, saving us nearly $2m in engineering hours.

https://nousresearch.com/refactoring-hermes-with-1393-agents

https://fxtwitter.com/NousResearch/status/2099984561451028913

Refactoring Hermes with 1,393 agents

Hermes Agent refactored its own codebase: 1,393 subagents over about nineteen active hours cut non-test Python by 34.4%, for roughly $19,300 in model spend against a $150k-$1.8M estimate for doing it by hand.

New blog post\:
︀︀
︀︀We had a million lines of Python to clean up\. On September 2nd @Teknium - https://x.com/Teknium asked Hermes Agent to do it\.
︀︀
︀︀1,393 subagents and nineteen hours later, the codebase was 34\.4% smaller, saving us nearly $2m in engineering hours\.
︀︀
︀︀nousresearch.com/refactoring-hermes-with-1393-agents - https://nousresearch.com/refactoring-hermes-with-1393-agents

💬 - https://x.com/intent/tweet?inreplyto=2099984561451028913 38 🔁 - https://x.com/intent/retweet?tweetid=2099984561451028913 33 ❤️ - https://x.com/intent/like?tweetid=2099984561451028913 409 👁️ 18\.0K 

Nous Research #announcements

@everyone

Announcing the Hermes Agent plugin catalog!

Hermes Agent now has a Plugin Catalog: starting with 4 official plugins and 96 from the community, covering desktop mods, new platforms, browsing, specialized tools, and more.

Our team reviews every community plugin, and we will add new ones regularly.

Want to submit your plugins? Learn how here: https://hermes-agent.nousresearch.com/docs/user-guide/features/plugin-catalog#submitting-a-plugin-to-the-catalog

https://fxtwitter.com/NousResearch/status/2100266421020152114

Plugin Catalog | Hermes Agent

Browse and install reviewed, SHA-pinned Hermes plugins from the curated catalog

Hermes Agent now has a Plugin Catalog\: starting with 4 official plugins and 96 from the community, covering desktop mods, new platforms, browsing, specialized tools, and more\.
︀︀
︀︀Our team reviews every community plugin, and we will add new ones regularly\.
︀︀
︀︀hermes-agent.nousresearch.com/docs/plugins - https://hermes-agent.nousresearch.com/docs/plugins

💬 - https://x.com/intent/tweet?inreplyto=2100266421020152114 2 ❤️ - https://x.com/intent/like?tweet_id=2100266421020152114 9 👁️ 2\.6K 

Nous Research #hermes-announcements

@hermes-agent-notifications

A bit of a roadmap:

https://fxtwitter.com/Teknium/status/2100645382552428963
https://fxtwitter.com/Teknium/status/2100645510856298543
https://fxtwitter.com/Teknium/status/2100645958447186331
https://fxtwitter.com/Teknium/status/2100646469015609779
https://fxtwitter.com/Teknium/status/2100646667259449653
https://fxtwitter.com/Teknium/status/2100647120147771645
https://fxtwitter.com/Teknium/status/2100647460217729401

We are going to lean into making Hermes more like Pi, and less like OpenClaw

💬 - https://x.com/intent/tweet?inreplyto=2100645382552428963 47 🔁 - https://x.com/intent/retweet?tweetid=2100645382552428963 9 ❤️ - https://x.com/intent/like?tweetid=2100645382552428963 249 👁️ 4\.1K 

↩ (@Teknium) - https://x.com/Teknium
The first step of this is removing all the bundled Memory providers\.
︀︀
︀︀They will be maintained by their creators, in their own org's repos, and land in the plugins market

💬 - https://x.com/intent/tweet?inreplyto=2100645510856298543 3 ❤️ - https://x.com/intent/like?tweet_id=2100645510856298543 55 👁️ 1\.2K 

↩ (@Teknium) - https://x.com/Teknium
This will be our test run of pulling integrations out of the core codebase, and into maintainer owned repos\.
︀︀
︀︀This means less to confuse users, less bloat, and less people who don't know what they're doing on a path that isn't the happy path\.

💬 - https://x.com/intent/tweet?inreplyto=2100645958447186331 3 ❤️ - https://x.com/intent/like?tweet_id=2100645958447186331 36 👁️ 751 

↩ (@Teknium) - https://x.com/Teknium
Now that we have a plugins catalog, and months ago made all integrations plugins that came bundled with Hermes, we can begin to move away from having to maintain and package them with every install\.
︀︀
︀︀This reduces our load as well, so we maintain the things that matter, while external integration creators maintain what matters to them

💬 - https://x.com/intent/tweet?inreplyto=2100646469015609779 1 ❤️ - https://x.com/intent/like?tweet_id=2100646469015609779 23 👁️ 550 

↩ (@Teknium) - https://x.com/Teknium
The biggest issue with independent plugins for everything was discoverability\.
︀︀
︀︀The Plugins API surface was already strong and capable of supporting all kinds of things, like tool providers, memory systems, etc \- but there was no discoverability until now\.

💬 - https://x.com/intent/tweet?inreplyto=2100646667259449653 1 ❤️ - https://x.com/intent/like?tweet_id=2100646667259449653 15 👁️ 528 

Nous Research #hermes-announcements

https://x.com/Teknium/status/2100647460217729401

Some confusion on what leaning into being more like Pi than Openclaw is being expressed \-

What I mean is, bundling and packaging less things with the core agent, more plugins, leaner core\. Not that we're not trying to be a personal assistant agent and becoming a coding agent\. For clarity\!

https://twitter.com/Teknium/status/2100647460217729401