Numinous (@numinous_ai) header
Numinous (@numinous_ai) profile picture

Mapping the drivers behind every event.

Bittensor Subnet #6
Joined November 2025
The competition will go live within 2 weeks. It won't be sandbox based anymore (we want miners to be able to leverage any kind of data) and will run in parallel to the current trajectory focused competition. Miners task will be to build our mappa mundi!
4
3
1
17
3,092
Cascade + Numinous = Bittensor forecasting world model
Track how the world changes. Predict how it evolves. We’ve partnered with @numinous_ai to combine knowledge graphs with time-series forecasting. Numinous maps relationships between companies and events, and predicts how those relationships may change. Cascade brings forecasting models to assist in flagging future activity, pointing agents towards worthy developments. In turn, Numinous’s graph will help us develop synthetic training data for our mixture-of-experts model, capturing how changes at one company can ripple through connected businesses. Together, we’re working towards better forecasts of both changing relationships and the numerical trends they influence.
3
1
17
1,145
Track how the world changes. Predict how it evolves. We’ve partnered with @numinous_ai to combine knowledge graphs with time-series forecasting. Numinous maps relationships between companies and events, and predicts how those relationships may change. Cascade brings forecasting models to assist in flagging future activity, pointing agents towards worthy developments. In turn, Numinous’s graph will help us develop synthetic training data for our mixture-of-experts model, capturing how changes at one company can ripple through connected businesses. Together, we’re working towards better forecasts of both changing relationships and the numerical trends they influence.
2
6
5
18
1,925
New leaderboard is out. Track our best agents.
2
4
1,345
Soon track our graph belief state; what models will be released? what supply chain will be disrupted?
1
2
352
The graph is adding 500 + relationships per day yet today the top miner only lists 6 correct statements. Join the competition and help push the forecast horizon. 👇 leaderboard.numinouslabs.io/…
2
137
Today starts our open-ended forecasting competition where miners forecast the future state of a dynamic knowledge graph. First focused on the top US equities and their supply chain. Tomorrow it will also have a geopolitics graph.
4
5
2
14
2,601
Every day we ask questions about entities in the graph such as NVIDIA or Broadcom. Miners answer a question with a bundle of one or more statements about a relationship that will appear in the news, for example "Paramount acquires Warner Bros". With the graph capturing 2 or 3 hop consequences of events, miners will forecast cascades and feed our world model.
1
5
1,073
Numinous retweeted
The competition will go live within 2 weeks. It won't be sandbox based anymore (we want miners to be able to leverage any kind of data) and will run in parallel to the current trajectory focused competition. Miners task will be to build our mappa mundi!
4
3
1
17
3,092
Which subnets win when gamma tokens go live? We mapped the subnets other subnets will actually pay to use: ⚡ Compute 🧠 Inference 📊 Data 🔮 Prediction 🔐 Security 📣 Marketing 🗄️ Storage 🤖 Agents …and more The full breakdown 👇
1
19
3
46
4,987
Numinous retweeted
it turns out that, with @numinous_ai pretraining, its quite good at crypto prices 👀
Publishing v0.57 on @huggingface now. This full version is the first Zeno model whose confident-failure warning beats a tuned gradient-boosted model on both held-out periods. That is the hardest comparator we run. What's new on it: • ONNX and safetensors files for all 6 trained copies of the model, plus the original PyTorch files. Every ONNX file gives the same answers as the original, with a difference under 0.0000005. • Official size: 9,160,161 parameters (9.16M). Design: 6 layers, width 320, 8 heads, plus a part that reads each forecaster's past record. Trained on @NVIDIA L40S. • Instructions and inputs vs outputs: a short ONNX example that runs on a normal computer, with no GPU or PyTorch needed. A table explains each input (the forecasters, their past records, how often pairs of them missed together, which source) and each output (the chance each forecaster is confidently wrong, and the correction to the forecast). There's also a backtest lookup tool and the hosted API/MCP. • Technical details file: the model's settings in the standard Hugging Face format, plus the exact training code and the run certificate.
1
3
21
772
Numinous-1 model is now publicly available! Both to miners and outside users. It is a SOTA predictive model in its weight class, beating other top open source forecasting models like DeepSeek and GLM. eversight.numinouslabs.io/do… It becomes the new subnet baseline. Miners will have to beat it to be rewarded. Reward concentrates. Level improves.
1
3
18
1,560
Today Numinous is releasing its impact UI! Built from our underlying causal graph, it shows for each equities their exposures along macro and geopolitics factors, tracked by prediction markets. As the event landscape changes, your fund can track the mechanisms propagating to equities.
7
8
2
37
3,902
The factors are publicly available on a restricted universe. Reach out to us to get access to our portfolio tool, factor matrix and supply chain focused factors! 👇 impact.numinouslabs.io/facto…
1
5
214
More coming soon on the subnet side to push agents to discover graph relationships!
5
203