Inference is beginning to dominate AI workloads. As demand for compute changes, so will the hardware best suited to run it.

Our network is a mix of hardware that can be matched to your workload. Read why that’s important here: https://blog.thetatoken.org/the-inference-era-why-the-next-chapter-of-ai-looks-different/

The new 2.0 release of AI Characters now supports offline, stateless sessions on EdgeCloud. AI characters stay in character without needing a continuously running session behind them, opening the door to text-based RPGs, interactive fiction, and many other types of games. Built on the same tools we launched with.

Find out how to build yours here: https://blog.thetatoken.org/build-ai-characters-into-your-game-with-theta/

Theta Labs has completed its annual treasury unstake, moving 30M THETA to support validator growth & network development into 2027.

The move also reduces Theta Labs' share of staked THETA, continuing the network's ongoing decentralisation.

Read more: https://blog.thetatoken.org/fueling-2026-2027-growth-annual-theta-validator-unstake/

GLM 5.3 and GLM 5.3 Flash by Z.ai are now available on Theta EdgeCloud.

Start building with them now: https://www.thetaedgecloud.com/dashboard/ai/service/on-demand-model-apis

Agents are beginning to dominate AI usage, consuming nearly 5x more tokens than humans. That presents an opportunity for us at Theta Labs. From today, our APIs and MCP server let AI agents find, deploy & manage GPUs on EdgeCloud on their own.

Read more: https://blog.thetatoken.org/ai-agents-can-now-deploy-edgecloud-gpus-themselves/

We didn't stop building over the summer, and there's a lot more in the pipeline before the end of the year.

Read more about what we got up to in August: https://blog.thetatoken.org/theta-roundup-august-2026/

Earlier this year, we announced a partnership with Syracuse University and AWS. Since then, Prof. Junzhe Zhang's team has been using Theta EdgeCloud Hybrid with AWS Trainium for ongoing research into generative AI, and we wanted to share a closer look at what they've been working on.

Generative AI models turn random noise into images by following a path. Zhang's team applied a principle from physics, the one that governs how objects move, to help models pick smarter, more suitable paths for the data they're working with.

Read more: https://blog.thetatoken.org/from-classical-mechanics-to-generative-ai-syracuse-university-advances-flow-matching-with-theta-edgecloud/

GLM-5.3 is now available on Theta EdgeCloud, both as an on-demand API and as a model option for our chatbots & customer service agents.

It's Z.ai's flagship model, built for coding and the kind of long-running agent work where a model has to hold a task together across many steps. GLM-5.3 uses the same base model as GLM-5.2, with every improvement coming from extended post-training rather than a new architecture. That produced a 50% gain on Z.ai's internal coding benchmark while using fewer output tokens than the version before it.

You can access the model now via Theta EdgeCloud: https://www.thetaedgecloud.com/dashboard/ai/service/on-demand-model-apis

Our CTO at Theta Labs, Jieyi Long, is the lead author on a paper out today that more than halves the estimated quantum cost of the core operation in breaking Bitcoin and Ethereum's cryptography.

Google Quantum AI published its estimates in March but kept the circuit private, attesting to it with a zero-knowledge proof. Eigen Labs took that verifier and made it an open challenge at ECDSA.Fail. Over two months, 100+ contributors working with AI agents beat the starting circuit by 86 percent, which puts it more than 50 percent below Google's own number. Google's researchers reviewed the paper.

No one has broken anything, and this is one subroutine rather than a full attack. But around a third of all bitcoin sits in addresses with exposed public keys, and migration takes years. These estimates are what the industry plans against, and they are now open and reproducible instead of locked behind a proof.

The co-authors included contributors from Eigen Labs, the Ethereum Foundation, StarkWare, Trail of Bits & more.

https://www.coindesk.com/tech/2026/09/10/crypto-researchers-cut-bitcoin-and-ethereum-quantum-attack-estimate-by-50

Last week, our CTO Jieyi Long led a paper with that more than halved the estimated quantum hardware needed to break Bitcoin and Ethereum.

We asked him a few questions about what it means and why it matters for the migration ahead.

Read more: https://blog.thetatoken.org/cto-jieyi-long-halving-bitcoin-quantum-attack-estimate/

Theta Labs and D.C. United have agreed to a multi-year partnership that puts a club-trained AI assistant in the hands of Black-and-Red supporters, answering their questions instantly, 24/7, and in the club's own voice.

Whether it is squad news before kickoff, directions to Audi Field on matchday or a question about tickets, supporters get an answer straight away. It is trained on club and MLS data, so what it tells you comes from D.C. United.

This is our third Major League Soccer deployment, after the San Jose Earthquakes and Philadelphia Union, and part of a wider group of clubs and leagues running fan-facing AI on Theta EdgeCloud.

Read more: https://www.dcunited.com/news/d-c-united-announces-partnership-with-theta-labs

🚀 Seedance 2.5 and BytePlus frontier AI models are now live on Theta EdgeCloud.

Developers can access ByteDance’s latest video, image, speech, and multimodal models, including Seedance 2.5 and Seedream 5.0 Pro, through Theta’s distributed AI infrastructure.

Read more:
https://blog.thetatoken.org/seedance-2-5-and-byteplus-frontier-models-are-now-live-on-theta-edgecloud/

Ling-3.0-flash-Fin is now available on Theta EdgeCloud through ZAN Router.

The model is built for finance-focused workloads, including company research, valuation, filings, reports and spreadsheet-based analysis. Its 256K context window supports work across large financial documents and multiple sources in a single workflow. 🧮

For developers, it is agent-native. Access through MCP and APIs makes it straightforward to use Ling-3.0-flash-Fin inside research agents, automated finance workflows and AI-native applications.

This also builds on our existing relationship with ZAN, which joined Theta Network earlier this year as a strategic Enterprise Validator.

Read more: https://blog.thetatoken.org/theta-edgecloud-integrates-ling-3-0-flash-fin-via-zan/

We’re excited to partner with Weaver Labs who will use our decentralized GPU network as an AI compute resource for Cell-Stack, its edge and connectivity platform for stadiums and venues, supporting low-latency inference for venue applications. 🏟⚡️

At Stadium MK, Cell-Stack delivered AI analytics, multi-angle streaming and 3D stadium navigation across the 30,400-seat venue. Venues want more of that intelligence running locally, and EdgeCloud is designed to give them elastic GPU capacity without heavy hardware investment.

Read more about the strategic partnership here: https://blog.thetatoken.org/theta-labs-and-weaver-labs-to-collaborate-on-ai-and-edge-infrastructure-for-stadiums-and-venues/

Alibaba Cloud, Asia Pacific's largest cloud provider is bringing Qwen, the world's most downloaded open model family, to Theta EdgeCloud. And we were thrilled to make it official in-person 🔥

In under a month, ByteDance's BytePlus, Ant Group's Ling and now Alibaba's Qwen have all chosen EdgeCloud to distribute their models.

Qwen3.8-Max and Qwen3.8-Flash are live now, on the API key you already use.

Read more: https://blog.thetatoken.org/theta-labs-announces-collaboration-with-alibaba-cloud/

🇰🇷 Kookmin University’s AI Vision Lab has joined Theta EdgeCloud’s academic network.

Led by Professor Youngwook Kim, the lab builds vision AI that learns reliably from imperfect data, for uses like satellite imagery, autonomous driving and medical imaging.

Real-world images are rarely fully labeled. A photo might show a person, a bike and a train, but only “person” is tagged. Treat the missing labels as “no” and the model learns wrong answers.

Prof. Kim’s CVPR 2022 paper “Large Loss Matters” showed those wrong labels stand out through large errors in training, so they can be caught and corrected. It beat prior state-of-the-art methods on major benchmarks.

Theta EdgeCloud’s decentralized GPU network gives labs like his affordable, on-demand compute to train on more data and iterate faster.

Read more 👇
https://blog.thetatoken.org/south-koreas-kookmin-university-ai-vision-lab-joins-theta-edgeclouds-academic-network/

AI agent traffic grew 7,851% last year. Agents now research, pay for services and deploy their own compute, no human in the loop.

Our Head of Marketing & PR on the autonomous web, and why blockchain laid the rails 👇
https://blog.thetatoken.org/the-web-is-starting-to-run-itself/

We’ve partnered with Q&SMART, the Seoul company behind DaVinciQ. 🇰🇷

DaVinciQ is an AI tool that helps students learn to think, not just find answers. Students pick a topic, debate it with the AI over five rounds, and get a report on their thinking skills with tips on what to practise next.

Each debate needs a lot of computing power, and that need grows as more schools sign up. Theta EdgeCloud provides affordable GPU computing that grows with them, so their team can focus on teaching.

Read more about what we're building together: https://blog.thetatoken.org/theta-edgecloud-partners-with-q-smart-to-power-davinciq-an-ai-debate-platform-built-for-critical-thinking/

ICYMI: Theta's October was another busy month with Alibaba Cloud bringing Qwen3.8 models to EdgeCloud, a collaboration with Weaver Labs to support AI workloads in stadiums, Kookmin University's AI Vision Lab joining our academic network, and the rise of autonomous AI agents.

https://blog.thetatoken.org/theta-roundup-october-2026/

We're excited to share the latest step forward for Theta EdgeCloud's AI gaming product! 🎮

In July 2026, we introduced AI Gaming and AI Characters, adding to our existing AI offerings alongside our GPU network/neocloud infrastructure and enterprise AI assistants and chatbots.

Today, we're launching our new AI Characters dashboard, making it easier than ever for developers to create, customize, and integrate intelligent, interactive characters into their games.

We're also introducing a new cost calculator to give developers greater transparency and control over their expenses, with straightforward pay-as-you-go pricing and no minimum spending requirements.

And you don't need to be a developer to try it! Chat with five unique AI characters or play our interactive poker demo to see AI-powered NPCs in action.

This is another important step for Theta EdgeCloud, and we have more exciting announcements coming over the next few weeks! 🚀

Explore AI Characters: https://www.thetaedgecloud.com/ai-characters