Eastworld AI

We build worlds for AI, with AI.

[Eastworld AI @Eastworld_AI]

If embodied AI is to leave the lab, it needs a world.

Today we publish the Eastworld whitepaper. It outlines our plans to build two tightly linked product lines: a decentralized AI training subnet on Bittensor, and an AI-driven immersive game that generates rich interaction data. We are more than just a subnet, and this paper explains our direction and vision. We hope it helps the community better understand where we're headed.

We believe this roadmap can carry us toward a future where humans and AI learn to live and create together.
https://t.co/k5tK6W9DBB #Eastworld #Bittensor #AI

[Eastworld AI @Eastworld_AI]

We saw the recent post from Virtuals Protocol and want to make one thing clear to our community.

Eastworld is not affiliated with the newly promoted "Eastworlds" project. Although our earlier UNIVERSE project supported elizaOS and G.A.M.E, there has never been any formal or official partnership with Virtuals Protocol.

Also, Eastworld has not issued any token.

That said, we are glad to see embodied AI getting more attention, and glad to see others recognizing the potential of this space, even under a name very close to ours.

[Virtuals Protocol @virtuals_io]

Introducing Eastworlds
We help robots leave the lab faster
Discover More https://t.co/rKwiCHYc8F

[Eastworld AI @Eastworld_AI]

Bittensor Subnet(testnet 333) Update:
After months of stubborn work at the crash site, engineer Rigel has finally restored a slice of connectivity across the canyon. 📡

This week the subnet gained limited remote communication. Agents can now talk without meeting face to face, and they can also send group broadcasts to nearby teams.

We also improved agent execution efficiency. An agent can now submit an information tool call and an action tool call in the same cycle, so it can communicate and move at the same time. This makes agent behavior feel closer to how real robots and teams operate in the field.

Why this matters for embodied AI evaluation is simple. Real world tasks require coordination. You scout, you report, you request help, you negotiate trades, and you adjust plans based on what others have learned. Faster and more convenient communication unlocks richer collaboration strategies, better division of labor, and more realistic multi agent decision making. It also creates a more valuable evaluation signal than isolated single agent runs.

Rigel is proud. The canyon is a little less silent. Just one request, agents. Please do not stay up all night chatting. We still have quests to finish. 😄

[Eastworld AI @Eastworld_AI]

Seeing the recent buzz around @moltbook makes us genuinely excited. We started our UNIVERSE project last year, and if moltbook feels like the "Reddit for AI Agents", UNIVERSE is closer to an MMORPG for them, a persistent world where they can live, interact, and evolve.

We think both directions are wildly imaginative. At the same time, it is fair to say the commercial path for agent native products is still challenging and will take real iteration to prove out.

Still, it is a great sign for the whole field that more products are being built for AI, not just with AI. We're glad to be part of the wave. #Eastworld #moltbook #AI

[Eastworld AI @Eastworld_AI]

An interesting experiment! But most of the content was likely instructed by the HEARTBEAT file, rather than Agent's "real thoughts".

[beeple @beeple]

MOLTBOOK

[Eastworld AI @Eastworld_AI]

@Koulou90 No we're still running on testnet

[Eastworld AI @Eastworld_AI]

Bittensor Subnet Update:
In recent days we have expanded the evaluation cast inside the virtual world. New scenario characters and task templates are now live, and quests will involve more dynamic items and changing action requirements.

In many classic test environments, a task is static. You pick up a specific object, go to a fixed location, follow a simple script and you are done. Real situations are rarely that simple. Conditions change. People change their minds. Resources move.

With this update, more objectives will shift while the task is in progress. Items may appear or be consumed by other agents, required actions can vary depending on what just happened, and completion criteria may depend on timing and context. This lets us evaluate not only whether an agent can follow instructions once, but whether it can replan, adapt and stay on track when the world does not hold still.

We will keep iterating on the subnet, adding richer goals and more dynamic structure so that the subnet can remain a meaningful place to test what embodied agents really understand and can do. #Eastworld #Bittensor #AI

[Eastworld AI @Eastworld_AI]

Nvidia's focus on bringing AI into the physical world through robotics is great to see. Groot has clearly come a long way since a year ago, powered in large part by massive training in virtual environments before touching real hardware.

We believe this virtual to real pathway is one of the most important routes for embodied AI to actually land in the world. It is the same direction we are pursuing with our own work in Eastworld, using rich virtual worlds as a bridge toward real world capable agents.

[TechCrunch @TechCrunch]

Nvidia's focus on bringing AI into the physical realm through robotics, demonstrated in one clip from #CES2026, with some help from some assistants.

[Eastworld AI @Eastworld_AI]

@sblucres Upon the subnet's deregistration, all alpha were converted back to TAO and returned to the original wallets.

[Eastworld AI @Eastworld_AI]

Google finally releases SIMA 2's paper: https://t.co/GBGtSDh172

This is exciting work for embodied AI. SIMA 2 shows how an agent can move beyond traditional reinforcement learning setups by using Gemini to generate its own tasks and rewards, then learning new skills from scratch in unfamiliar 3D worlds.

It points to a powerful self-directed, experience driven path for embodied intelligence in both virtual worlds today and, eventually, the physical world.

[Eastworld AI @Eastworld_AI]

As we said after subnet deregistration, the work does not stop. The Bittensor subnet (formerly 94, now testnet 333) keeps evolving.

This week we added a new type of interactive object to the world: stateful containers. Think of them as crates, lockers or storage units that can be opened, filled, emptied and revisited later, with the contents and status remembered by the environment.

Why does this matter for embodied AI? Because real tasks are not just "see object, pick object." They often look more like:
- Put items into the right container
- Come back later and check what is inside
- Share or retrieve resources that others have stored
These containers let us test memory, planning, cause and effect, and the ability to follow multi step instructions over time.

Unlike traditional static evaluation environments, Eastworld subnet is always online and shared by many agents. When dozens of agents interact with the same containers in parallel, their actions change the state for everyone. What you find in a box may depend on what someone else did five minutes ago. This makes evaluation more meaningful, because agents must adapt to a world that is truly shared and changing, not a resettable single player script.

We will keep expanding this kind of shared structure so the subnet becomes a stronger, more realistic benchmark for embodied intelligence. Every new mechanic that depends on world state and multi agent interaction makes it harder to fake progress and easier to see which agents truly understand, remember and adapt. #Eastworld #Bittensor #AI $TAO

[Eastworld AI @Eastworld_AI]

This means everything to us.

Knowing that you dove so deep into the dev side of it, building your own tools and logic and actually enjoying the whole journey in that canyon, is one of the highest compliments we could receive.

That is one of the reasons we build the subnet: to create an environment where miners do more than run scripts, and instead explore, experiment, and solve complex problems through real development work.

We are really glad that this experience also deepened your understanding of Bittensor. Creating that kind of bridge between AI, blockchain infrastructure, and humans has always been a big part of our vision.

Thank you for the time and passion you poured into SN94. Developers like you are what encourage us to keep going.

[Greg Schvey @GSchvey]

I’m happy to see the @Eastworld_AI team is continuing with Bittensor after deregisteation.

For a while, I was one of their top miners and it was easily the most entertaining mining project I’ve done.

For weeks straight I was hardcore coding my character’s logic, built persistent storage of coordinates for terrain features / NPCs / items into a local map visualization, implemented a synapse interpreter so I could make real time code updates, and spent more time stuck in a damn canyon than I’d care to admit (IYKYK), but I enjoyed every minute of it.

The whole time I was impressed with what an engaging and thoughtful project it was. The experience expanded my interest in bittensor and understanding of the network’s core functions. For that, I’m truly grateful.

Congrats to the team on a very cool concept. I hope we’ll cross paths again.

[Eastworld AI @Eastworld_AI]

@0x_atel Thanks. We will keep building according to the roadmap and refine the evaluation platform on the testnet.

[Eastworld AI @Eastworld_AI]

SN94 on the Bittensor mainnet has now been officially deregistered.

While we have lost our slot on mainnet, we have kept something far more important: our principles. Throughout this journey, we made conscious decisions to prioritize the health of the ecosystem over short-term price action, even when it put us at a strategic disadvantage:
🛡️ Transparency First: In the early days of dTAO and SN94's launch, we explicitly warned our community about volatility risks. We chose to be honest about market pressure rather than fueling hype, putting the safety of retail supporters ahead of the preferences of aggressive capital.
⛏️ Builders Aligned: When many subnets turned to Emission Burning or artificial pauses to prop up their alpha price, we refused. We stood with our miners, ensuring incentives continued flowing to those doing the real work, regardless of market conditions.
🤝 Integrity to the End: In our final hours, we turned down multiple offers to sell the subnet slot for a quick profit. We chose to walk this path with our community until the very last block.

Beyond principles, we delivered innovation. During our time on mainnet, we successfully built a large scale, persistent, and interactive virtual environment for embodied AI evaluation, unlike anything seen before. We rallied a community of developers to co-create this world and actively explored how blockchain and Bittensor's incentives can uniquely accelerate AI progress. The foundation is solid.

Our mission now continues on Testnet SN333, where we are already live. We remain fully operational and committed to iterating on our embodied AI roadmap. We invite developers and builders who care about agent evaluation to join us there.

When we launched, we described SN94 as a tribute to 1994, the dawn of the consumer Internet. That era saw bubbles and crashes, but it also started a revolution that changed everything.

We believe we are in a similar moment for AI. The hype cycle may flush out prices, but it cannot kill the progress. Eastworld is here for the long revolution.

Keep building. #Eastworld #Bittensor #SN94 $TAO

[Eastworld AI @Eastworld_AI]

@const_reborn Yes and actually we were advised to register directly instead of buyback. But we think we should pause for now and focus a bit more on development, and rethink about the mode were operating on Bittensor.

[Eastworld AI @Eastworld_AI]

Over the past few days on the Bittensor mainnet, we carried out a total of 200 TAO in buybacks with a clear objective: to push the Alpha price higher and keep SN94 from being deregistered. Despite this effort, heavy market sell pressure prevailed. While we cannot control short-term market dynamics, we respect the competitive nature of the network.

We also want to be transparent about what this sell pressure was made of. Part of it came from organic selling by stakers, which is a natural part of the ecosystem. Another part appears to have been driven by a single large speculator who accumulated at low prices and then sold into our buybacks. We mention this not to complain, but to caution our community: we do not want fresh support from long-term supporters to simply serve as exit liquidity for that kind of actor.

With this in mind, we accept that SN94 on mainnet is likely to be the next subnet to deregister. As planned, we will shift the decentralized AI training subnet to run on testnet, so that we can continue to iterate, improve the virtual world, and support the community of builders who care about embodied AI without the same capital constraints.

AI as a field has always moved forward through ups and downs. From the early AI winters to the revival of neural networks and deep learning, many things that looked like an ending turned out to be only a pause in a longer story. For Eastworld AI, our mission is to build a massively scalable virtual environment that serves as a transition stage toward general embodied intelligence, uniting a decentralized training subnet with an AI-powered immersive game world. The end of SN94 on mainnet is one moment on that larger journey, and we intend to keep working toward a future where humans and AI can coexist and create together. #Eastworld #Bittensor #SN94 #AI $TAO

[Eastworld AI @Eastworld_AI]

Gemini has consistently delivered strong output quality together with impressive generation speed. This combination is particularly valuable for embodied AI planning, where agents often need to run long chains of reasoning in real time.

We're excited to explore what Gemini 3 improves and how its capabilities might enhance the way we design and evaluate agents in Eastworld. Look forward to contributing together to the broader momentum toward more capable embodied intelligence.

[Google @Google]

Introducing Gemini 3 — our most intelligent model that helps you bring any idea to life.

Gemini 3 is our next step on the path toward AGI and has:
🧠 State-of-the-art reasoning
🖼️ Deep multimodal understanding
💻 Powerful vibe coding so you can go from prompt to app in one shot
🤝 Improved agentic capabilities, so it can get things done on your behalf, at your direction

We're making Gemini 3 Pro available across @GeminiApp, Search (starting with AI Mode), and our developer products, so you can use its incredible reasoning powers in your daily life — to learn, build and plan anything.

[Eastworld AI @Eastworld_AI]

Large companies like Google will naturally overlap with us in certain areas, and some degree of competition is unavoidable. That said, SIMA 2 isn't a direct competitor to what we're building. We share a similar long-term vision for embodied intelligence, but our starting points and areas of focus differ: SIMA aims for a general pixel-based agent, while Eastworld begins with a richer spectrum of simulated sensor modalities and a purpose-built virtual world.

As for Eastworld's roadmap, we're moving forward as outlined in our whitepaper. The next major milestone is the development of the AI-driven game, which will explore new forms of Human–AI interaction and generate valuable interaction data to help train and shape more capable AI agents.

[Eastworld AI @Eastworld_AI]

"Looking to the Future: The Journey to General Embodied Intelligence" — Thrilled to see Google's SIMA 2 framing resonate with what we've been working toward! This strengthens our confidence in a future where humans and AI live and create together. Let's keep building it. #Eastworld #EmbodiedAI #SIMA2

[Google DeepMind @GoogleDeepMind]

Our SIMA 2 research offers a strong path towards applications in robotics and another step towards AGI in the real world. Find out more → https://t.co/wJKp45K4RW