TAO.com

“I got promoted to a pro account today so I had multiple questions, I just went to their discord to create a ticket and they replied so fast all my questions were answered on point, Great customer support fr, and also I have received 11 payouts (still counting) from this firm in case you are wondering if they are trustworthy. Thank you Vanta” - From Oladeinde on Trustpilot This is why we built Vanta - to help traders succeed, grow, and get paid. We call it the Vanta Difference.

Now Live: Beam’s @Box integration. Box has 97,000 customers and supports 68% of Fortune 500 companies. This expands Beam's programmable bandwidth network even further.

An open model now sits a tenth of a point behind Fable 5 at coding in a terminal. That's DeepSeek V4 Pro on Terminal-Bench 2.1: 87.9 against Fable 5's 88.0, with Opus 4.8 at 85.0 (DeepSeek's published scores). What to use it for: - a coding agent that runs commands, reads the errors and fixes what broke - security review, where it tops DeepSeek's Cybergym table at 83.3 Its 1M-token context lets you hand it a big slice of your repo at once, so the agent sees how the pieces connect instead of guessing file by file. It's weaker on long, multi-step projects (62.7 on DeepSWE against Fable 5's 70.0), so try it on your own tasks before trusting it with those. On Chutes it's $0.85 in and $2.65 out per million tokens, at any hour. Model page and pricing: chutes.ai/app/chute/chutes-d…

October on SN44, in a few lines. -More public tracking tasks running on the subnet -3 new public elements out -A big chunk of example data from the cricket private track is now public -A lot of quiet work on security and scale -We can spin up new tasks faster now, with plenty more coming And following IBC, one of the largest real-time graphics software companies is now testing a Studio endpoint, running on the football model built earlier this year under our previous incentive mechanism. Meanwhile, we’re collecting feedback from Studio users for what comes next. From here the pace only picks up, with way more tasks every month and one goal, making computer vision a commodity.

NOVA: Emission Split Update for Data Generation Fund Owner 26% / Miners 33% / Validators 41% With the launch of eTAO, subnet owners can now adjust how emissions are distributed. NOVA will be updating its split to allocate more resources toward what we believe is the highest-impact investment we can make in the subnet's future: experimental data generation. Wet lab data is the best investment we can make both to prove out what the subnet produces and to increase the value of what gets built on top of it. Data is the bottleneck AI-driven drug discovery faces a fundamental constraint: the quality of our predictions is ultimately limited by the experimental data available to train and validate them. Much of today's publicly available drug discovery data is fragmented across studies, generated using different experimental protocols, and inconsistent in quality and coverage. Generating high-quality experimental data allows us to address that limitation directly. It gives miners better ground truth for model development, provides independent validation of their predictions, and helps us identify which approaches actually translate into experimental success. The industry is making the same bet. Some of the largest investments in AI-driven biology are now directed toward generating the experimental data necessary to train and validate better models. Biohub committed $400M of a $500M, five-year initiative to data generation, infrastructure and measurement technology, and later expanded into a $1.8B effort backed by Meta, Google DeepMind and Isomorphic Labs, who collectively committed $300M. The OpenAI Foundation gave $15M to OpenADMET to generate public ADME datasets, which will be used in prospective blind challenges against newly measured experimental data. Octant continues the high-throughput data generation behind it. Leash Bio used a $9.3M seed round to scale data collection for a foundational medicinal chemistry model, producing a dataset that screened 133 million molecules against each protein. Anthropic has set up a physical biology lab in the Bay Area. What the additional funds will support The additional owner allocation will go into a dedicated R&D fund for wet lab data generation. 1. ADME model training competition. The first use of the fund is generating the experimental data behind our ADME competition, currently in design and development. High-quality, standardized ADME data is what turns a competition from a benchmark into something that produces models with real predictive value. 2. Advancing assets under development. Where it makes sense, the fund can also support experimental work that moves our in-development assets forward. Why this benefits the entire NOVA ecosystem We believe generating experimental data is one of the most effective ways to increase the long-term value of the subnet. For miners, better datasets and scientifically grounded evaluations create opportunities to develop more capable models, compete on meaningful performance improvements, and contribute work with greater real-world utility. For validators, stronger experimental benchmarks improve the ability to assess model quality and reward useful contributions. For tokenholders, better models, experimentally validated discoveries, and progress toward valuable therapeutic assets can strengthen NOVA's competitive position and long-term commercial potential. The economic opportunity extends beyond any individual competition. We are building an increasingly capable scientific discovery platform, and the data generated through our R&D efforts can become a compounding asset for the entire ecosystem. Miner incentives remain strong Miners will still receive a meaningfully larger share than the owner (33% vs. 26%), and we believe the rewards will preserve healthy, competitive incentives for participation. Better data and better-validated competitions also make the work miners do more valuable over time. Protecting token health Funds from the R&D allocation will be converted to cover lab costs gradually, distributed over time rather than sold in size, to limit market impact and preserve token health. The bottom line We think this is the single best thing we can do to accelerate our progress, increase our chance of success, and create value for everyone with a stake in NOVA, including miners, validators, and tokenholders. It is time for us to scale real-world validation.

The Franka Tabletop RL Challenge, Phase 1, is now live on SN49. 🚀 Miners build the brain that teaches a Franka arm to grasp, carry, place, stack and pack objects on a cluttered table, creating high-quality practice data for next-gen AI models. The robot can see the objects around it. Everything else stays fixed: the table, the physics, the setup. We give you a starting point, and you can change anything. A task only counts if the robot holds its final position for half a second. The top 3 split the rewards 70/20/10. Submissions close Sat 14 Nov.

Every company shipping AI agents asks one question: how can I stop my agent from going rogue? To answer it, HaloGuard Output Guard now sets a new state of the art (SOTA): 83.63 average F1, ahead of the previous best at 82.5. All measured on public benchmarks. And we did it with a 0.8B model, a quarter of the size of the previous best. Small enough to run efficiently. Cheap enough to make checking every response practical. This was the focus of Phase 3 on Trishool: • 7,500+ jailbreaks submitted • 125+ unique miners • 15 weeks of challenges It comes at the right time. We’re taking our product to enterprises now, and this makes our security stack stronger. We’ll share the models on Hugging Face, the arXiv paper and a full explainer soon. There’s much more to come, and we’re looking forward to an exciting week ahead. SOTA AI security, on Bittensor. Ψ

Miners with published forecasts fell from 235 to 43 between the Oct 21 and Oct 22 rounds, following a protocol update on our side. The consensus score dipped to 94.4% from 96.0%, still inside the 93.6% to 97.1% range it has held since Oct 4. Oct 18 to 24 averaged 95.1% over 420 scored values, with no missing consensus in any round. Three rounds in, a much smaller miner set is producing results in the same range. It’s the first big swing in miner count since the early rounds, so we’ll see how the score settles.

Mistral 4 Large live on SayGM. Mistral released the weights yesterday. You can use it on SayGM today. It's Mistral's new open-weight flagship, served by @KubeTEEAI inside a TEE, so your prompts stay sealed end to end.

New: Validate a quant tool. With TPN Bench quant verification tool, users can deterministically or probabilistically prove that a quantization is indeed a quant of the base model. The tool's report shows how often the quant picks the same next word (top-1 match), how far its predictions drift (KLD, KL divergence) and how much worse it predicts text (perplexity) compared to the parent model. This is something that we will definitely use internally, but it is also a tool that devs and labs might use if they have a quantization, in order to verify their quants independently. Now live on bench.trueperformancenetwork…

SN105 Beam is now live on Hugging Face, and its Zapier integration is live too. Teams can move datasets and model weights in and out of HF through a subnet, or make it one step in workflows they already run. Demand starts where users already are.

Now Live: Beam + @Zapier. 9,000+ apps. Over 2 million businesses are automating with Zapier. Picture those everyday workflows with Beam’s data movement built in. An app event triggers a transfer. Your data moves where it’s needed. The workflow keeps going.

Since June 17th, we've hashed every single Zeus forecast to the blockchain. Now, 25K times our historical forecast archive has been downloaded on HuggingFace. Our trustless verification mechanism allows anyone to check that a published forecast matches the hash commited on-chain at the recorded block. 🔗 huggingface.co/datasets/orph…

DeepSeek V4 Pro is live on Chutes. It's the big one from DeepSeek's V4 family, with 1.6T parameters and a 1M-token context window. Open weights, MIT licence. It's strong on agent work. DeepSeek's published score on Terminal-Bench 2.1 is 87.9, ahead of Opus 4.8 at 85.0. $0.85 per million input tokens, $2.65 out. Try it here: chutes.ai/app/chute/chutes-d…

SN67 miners can now call @OpenRouter rerank models as a tool. Search and fetch still find the pages. Rerank sorts the candidates before you burn an llm_chat call on the wrong ones. Paid models only. Same OpenRouter key. $TAO #SN67

27 people trained policies for one job: get a Unitree Go2 through a maze it has never seen, using LiDAR only. The winner scored 0.9272 on held-out mazes. Phase 1 had a smooth, clean floor. Phase 2 is live now on rugged, uneven terrain. #Robotics @opentensor @bittensor @NVIDIARobotics

Now Live: Beam💫 x @huggingface🤗. Nearly 3 million public models. 1 million datasets. Hugging Face has become a major hub for open AI, with its dataset count growing from 711,000 to 1 million in the first seven months of 2026. That scale creates a practical challenge: getting massive datasets to the compute that needs them. Beam will enable teams to move large datasets into and out of Hugging Face through its distributed network, connecting the data behind AI to the infrastructure that puts it to work.

Synth now ties miner rewards to collective performance. AutoQuant uses models built by frontier AI agents to set a benchmark: the further miners outperform it, the less is burned and the more goes to miners. This week’s miner burn: 22%, unchanged from last week.