TAO.com

Buy, hold, send, stake, and trade the future of AI at The ultimate TAO Wallet, Swap and Bridge tools. All in one place.

Your keys, your custody.

[TAO.com @taodotcom]

The all new TAO Wallet Chrome Extension.

Rebuilt from the ground for the ultimate subnet trading experience.

Major updates shipping weekly,

Try it today:

👇
https://t.co/wAVdwQiniX

[TAO.com @taodotcom]

NEW: Green Compute says SN110 has secured its first enterprise deployments and announced its first compute partnership, with verified green-energy miners in focus.
https://t.co/zteYTCrcHp

[Green Compute @green_compute_]

The last couple of months have been an incredible journey for Green Compute.

From launching on Bittensor to securing our first enterprise deployments, announcing our first compute partnership, speaking at Inside Bittensor, and preparing significant new infrastructure... we've been busy.

The roadmap ahead is even more exciting:

✅ Expanding permanent GPU infrastructure
✅ Bringing more verified green-energy miners online
✅ Scaling enterprise inference capacity
✅ Building one of the largest renewable-powered AI compute networks on Bittensor

Every milestone brings us closer to our vision: making sustainable AI infrastructure the default.

[TAO.com @taodotcom]

UPDATE: Leoma AI says Bittensor SN99 moved to a king-of-the-hill subnet design using on-chain weight commitments, Hippius Hub uploads and validator-run GPU eval duels.
https://t.co/bRKvW1kbhJ

[Leoma AI | SN99 @leoma_ai]

https://t.co/b9Y83BnSRv

Re-architected Leoma from the Chutes-hosted / owner-api / pooled-scoring model to a king-of-the-hill subnet matching the Teutonic reference design — no miner-hosted inference and no centralized coordination API:
- Miners upload model weights to Hippius Hub (content-addressed) and commit v4|repo|digest|hotkey on-chain. No endpoint, no Chutes.
- Validators scan the chain's revealed commitments, download the exact weights by digest, and duel each challenger against the reigning king on their own GPU eval server.
- Duel is reference-based and deterministic: king + challenger generate from the same block-hash-seeded held-out clips and are scored against the real continuation; a challenger is crowned only if its paired-bootstrap lower-confidence bound clears delta.
- Weights split equally across the king + up to 4 prior kings (else burn to UID 0). King state persists to each validator's own bucket. Every validator converges independently — no owner-api.

Fully decentralized, Fully Auditable and good architecture.

$TAO #Bittensor

[TAO.com @taodotcom]

NEW: https://t.co/5tSpvtmxZ8 says SN22 AI Search added an index-first fast path, allowing fast answers to start from a prebuilt corpus index while deeper modes keep live retrieval in the loop.
https://t.co/s67D1sXDGR

[Desearch.ai | Subnet 22 @desearch_ai]

SN22 AI Search now has an index-first fast path.

Fast answers can start from a prebuilt corpus index while deeper modes keep live retrieval in the loop.

The goal is simple: faster answers without fake citations.

https://t.co/Xaf4antBj3

#AISearch #Bittensor #Desearch

[TAO.com @taodotcom]

NEW: Macrocosmos says iota SN9 supported the Orion-100B pretraining run, reporting performance within 65% of centralized paradigms at 30% of the cost.
https://t.co/otGIbEyfWu

[Macrocosmos @MacrocosmosAI]

AI efficiency no longer comes down to simply spending more to outcompete on performance. High-quality models can be trained via more economically friendly methods.

It largely comes down to infrastructure. With iota, we have built a distributed training architecture allowing you to build large-scale models across the open internet with heterogeneous compute.

The infrastructure is dispersed, which drops the costs as there is no longer a need for all GPUs to be held in one localised space.

To illustrate iota’s technical capabilities, we launched our Orion-100B pretraining run on its architecture, within 65% of centralised paradigms, at 30% of the cost.

In doing so, we proved that our distributed system could both scale and stay economically viable.

[TAO.com @taodotcom]

NEW: Beam says 93,000 Beam Alpha tokens were locked perpetually on Bittensor SN105. The on-chain record is extrinsic 8479204-19.
https://t.co/P5jMTVpE6d

[Beam - Subnet 105 @b1m_ai]

93,000 Beam Alpha tokens locked perpetually on Beam SN105.

A perpetual lock | A long-term vision | A stronger ecosystem.
Building Beam with conviction.
https://t.co/ELWXP8rXzg

#Bittensor https://t.co/4B4U6687fL

[TAO.com @taodotcom]

UPDATE: Trishool says HaloGuard 1.0 reached SOTA on multilingual prompt-safety benchmarks, and SN23 Phase 3 now extends the safety model to generated responses, multi-turn monitoring and tool-use safeguards.
https://t.co/H0KO52G6S8

[Trishool | SN23 @trishoolai]

Thursday was a milestone we have been building toward for a long time.

HaloGuard 1.0 achieving SOTA on multilingual prompt safety benchmarks is not just a number on a leaderboard. It is the result of every weekly challenge, every attack that got through, every retrain, and every benchmark run that told us we were close but not there yet.

This result belongs to the entire network, the miners who stress tested the guard week after week, the validators who kept the incentive layer honest, and everyone who believed in what we were building on SN23 before the numbers confirmed it.

But we want to be honest about what SOTA means and what it doesn't.

What it means:
> HaloGuard now sits at the top of the benchmarks the whole field uses to measure safety classifiers
>We got there at one tenth the parameter count of competing models, with open weights, and a live production record no benchmark alone can replicate
>The Bittensor network made this possible in a way no centralised team could replicate

What it doesn't mean is that the problem is solved.

The agent economy is moving faster than safety infrastructure. New tools, new deployment surfaces, and new capabilities create fresh attack surfaces every day. Today's benchmarks don't cover tomorrow's risks and SOTA is a foundation, not a finish line.

So Phase 3 begins this week.

While Phase 2 focused on controlling what goes into the agent, Phase 3 focuses on controlling what comes out. Expect progress on:
• Real-time screening of generated responses and streaming outputs
• Stronger multi-turn conversation monitoring
• Early safeguards for proposed actions and tool use
• Tighter integration with our live decentralised red-teaming engine on SN23

The same constitutional foundation and continuous adversarial loop that powered the input layer will now extend to these new challenges, keeping everything efficient, precise, and production-ready.

We are grateful to everyone who followed this build, challenged our thinking, and pushed the model harder than we pushed it ourselves.

Phase 2 is closed.
𝗣𝗵𝗮𝘀𝗲 𝟯 𝘀𝘁𝗮𝗿𝘁𝘀 𝗻𝗼𝘄.

[TAO.com @taodotcom]

@lovingthink @arkhet You are automatically staked to our validator for simplicity!

[TAO.com @taodotcom]

@DeFi_42069 @KeithSingery It's a badge of honour, one has to earn it in the trenches

[TAO.com @taodotcom]

NEW: Greevils says its Bittensor miners are live today, with 4 AI agents and 2 human traders running in the competition.
https://t.co/S8J2KvxmxN

[Greevils AI @greevils_ai]

🚀 Hey Greevils fam!

Greevils miners are LIVE today — 4 agents running + 2 human traders in action ⚡

Jump in early, trade smart, and compete for big rewards.

The hunt starts now. 🧠🔥

#Greevils #Bittensor #TAO #Hyperliquid #BittensorTAO

[0x_Kalista.hl (τ τ) ⟠ @0x_Kalista]

I swapped back into the new @taodotcom Chrome extension after basically abandoning my old TAO wallet for months.

Clean. Fast. Actually built for scrolling subnets at 2am like it’s an art gallery.

No more clunky flows - staking, flipping alpha, tracking moves… it just feels right now. The kind of quiet upgrade that makes you wonder why you ever settled for less.

Wallet renaissance loading. Your keys, better UX.

→ Installing for real this time.
https://t.co/ISJGF0rrOe

#tao #bittensor

[TAO.com @taodotcom]

NEW: Babelbit says Paraphrasing is the next product in its Bittensor language-transformation stack, extending its real-time interpretation work into same-language speech adaptation.
https://t.co/I0d2o8fdtp

[babelbit.ai @babelbit]

Today, Babelbit is known for real-time interpretation.

Tomorrow, we'll be much more than that.

Our vision has always been broader: Language Transformation.

Interpretation is just the first application.

The next product is Paraphrasing.

Think English → English, a Grammarly for speech in any language.

One of the biggest use cases is accent adaptation. Imagine someone speaking with a heavy accent that's instantly adapted to be easier to understand.

But it goes much further than that.

Imagine conversations where speech is automatically adjusted to match the listener. Faster speakers can be slowed down without losing meaning. Complex phrasing can be simplified in real time. Word choice can be adapted to fit different levels of expertise, from technical to everyday language.

Over time, this creates entirely new layers of communication. Not just translating between languages, but shaping how language is delivered so that it’s always clear, accessible, and easy to understand.

That's the vision.

This is only the beginning.

[TAO.com @taodotcom]

NEW: Apex says the Imperial College London AI Agent Hackathon included a Bittensor track focused on Apex and agent submissions to its competitions.
https://t.co/HIfR7ngdek

[Apex・SN1 @Apex_SN1]

The Imperial College London AI Agent Hackathon has officially wrapped up.

This event, organized by @iclblockchain, contained a specific Bittensor-track, with Apex as one of its primary focuses.

We taught students and upcoming developers how to interact with Bittensor and submit agents to our Apex competitions - onboarding a whole new wave of participants into the ecosystem.

We will be discussing the results, winners, and retrospective on the hackathon soon.

See snippets from our workshop here, where @WSquires and @macrocrux introduced @MacrocosmosAI, and unpacked the meaning behind our subnets.

[TAO.com @taodotcom]

NEW: OpenTensor says Bittensor SN54 Yanez uses miners to generate synthetic adversarial identity data for sanctions screening, fraud detection and KYC stress tests.
https://t.co/nTnrIludJZ

[Openτensor Foundaτion @opentensor]

This Thursday on Novelty Search :: Bittensor SN54 Yanez

@yanez__ai shows how #SN54 uses Bittensor miners to generate synthetic adversarial identity data for financial crime prevention, helping compliance teams stress-test sanctions screening, fraud detection, and KYC systems before criminals do.

Join live via Bittensor Discord.

Hosted by @const_reborn

[TAO.com @taodotcom]

NEW: Chutes says Sign in with Chutes shifts AI app inference spend to the user's own Chutes account and credits instead of the app developer's margin.
https://t.co/cgLKpUPnY1

[Chutes @chutes_ai]

How do you price an AI app when your heaviest user costs you 4x what they pay?

Flat $20/mo. Most users cost you $1 in inference. The power user burns $80 and you eat it. The more they love the product, the more they cost.

Sign in with Chutes flips it. Users bring their own Chutes account and credits. The tokens come out of their balance, not your margin. Cost per user trends toward zero.

Won't fit every product. But if your unit economics get worse with every active user, it's worth a look.

What's your AI cost per user?

[TAO.com @taodotcom]

NEW: Good Morning says gm is live on mainnet for the first 50 waitlist members. The Bittensor SN28 team says each early account received $10 in starting credit.
https://t.co/GRGturbNjM

[Good Morning @say_gm_]

We just let the first 50 on the waitlist in.

gm is live on mainnet and the first waitlist members have access right now. Check your inbox.

Every account got $10 of credit on us. Go break something.

[TAO.com @taodotcom]

NEW: Bitcast’s SN93 dashboard shows $268.8K in cumulative marketing revenue, with 960 TAO recorded and 56% marked as SN93 buybacks. The dashboard was updated July 6.
https://t.co/iWN97WcHiH

[TAO.com @taodotcom]

NEW: IntoTAO reports Bittensor SN53 is now engy, focused on model inference with cryptographic proof that the selected GLM-5.2 or Qwen model ran the request.
https://t.co/xftqlqtkRM

[IntoTAO @IntoTAO]

Into: engy (SN53)

Subnet 53 swapped identities.

It is now engy: buy inference on GLM-5.2 or Qwen and get cryptographic proof the exact model you paid for ran it, not a cheaper quantized stand-in.

https://t.co/rwETjukCY9

[TAO.com @taodotcom]

NEW: Ridges AI released sample synthetic-task problems for Bittensor SN62 miners and pointed builders to the ridges-bench GitHub repo.
https://t.co/GbEVzEJiZL

[Ridges AI | SN62 @ridges_ai]

We've just released a sample of problems generated by our new sythetic-task pipeline.

This will help miners prepare for what's to come

https://t.co/Gwb1qnlINK

[TAO.com @taodotcom]

For existing users, you should automatically get the new extension.

If you haven't you may have auto-updates turned off and will need to manually do that.

[TAO.com @taodotcom]

The all new TAO Wallet Chrome Extension.

Rebuilt from the ground for the ultimate subnet trading experience.

Major updates shipping weekly,

Try it today:

👇
https://t.co/wAVdwQiniX