Keo

[Keo @wktob]

@archerlin__ Thanks

[Keo @wktob]

@archerlin__ Hi, done. Thanks

[const @const_reborn]

root is coming back, and it's going to be glorious

[Keo @wktob]

@fish_datura Thanks for supporting alpha holders.

[Z.ai @Zai_org]

The "triple usage" period for GLM-5.1 and GLM-5-Turbo is now extended to June 30.

Availability: Anytime except 2-6 AM ET.

[Keo @wktob]

@fish_datura 👍

[Fish | Datura @fish_datura]

Bittensor always adapts and evolves from events like this. The result is always an improved and more resiliant protocol.

We will lock our owner tokens in lium once this is implemented, in addition to continuing to burn our owner cut.

[Fish | Datura @fish_datura]

I think this should be the new guiding principle of subnets:

Find an evaluation metric that correlates to a performance metric whereas overfitting to the performance metric does not cause an increase in the evaluation metric, but overfitting to the evaluation metric does cause an increase in the performance metric.

Hopefully we see more subnets that correctly implement this strategy.

[Fish | Datura @fish_datura]

Crazy how a bittensor subnet has already trained a model better than qwen's own 4b model within weeks of being live.

The subnet doesn't even incentivize benchmark scores, proving that these benchmark scores are an effect of genuine improvements to the model as opposed to overfitting.

read full eval paper: https://t.co/qaqsftblR3

[Fish | Datura @fish_datura]

In the midst of massive GPU shortages, every useful GPU type is available on lium now. More than most other GPU providers. The power of bittensor and incentive mechanisms.

[Fish | Datura @fish_datura]

Be careful of subnets promising buybacks and not delivering.

Be careful of subnets not burning their buybacks.

Be careful of subnets that are not transparent about their revenue or don't have a clear plan to support the alpha holders.

Let's learn from what happened with covenant. DYOR!

[Fish | Datura @fish_datura]

"Bittensor isn't decentralized"

Miners from all subnets look like this. Working permissionlessly all across the world toward a common goal.

From the people, for the people, by the people. No single point of failure.

[Chutes @chutes_ai]

A 27B dense model outscored Claude 4.5 Opus on vision.

Qwen3.6-27B. Live on Chutes.

Qwen's newest open source flagship. Image, video, and text native. 262K context (extensible to 1M). Apache 2.0.

Head-to-head vs Claude 4.5 Opus on vision:
- V*: 94.7 vs 67.0
- CountBench: 97.8 vs 90.6
- VideoMME (w/ sub): 87.7 vs 77.7
- ERQA: 62.5 vs 46.8
- CharXiv RQ: 78.4 vs 68.5

Terminal-Bench 2.0 matches Claude 4.5 Opus at 59.3. New "Preserve Thinking" mode keeps reasoning traces across turns for agent workflows.

Running inside a TEE on Chutes. The GPU operators serving the model can't see your prompts or outputs.

$0.195 in / $1.56 out per million tokens.

Try it: https://t.co/labWgerqnq

[Keo @wktob]

"Top holders" doesn't seem to accurate. Anyone experiencing this?

[Keo @wktob]

Is https://t.co/9sZnSZYG6a down?

[Jon Durbin @jon_durbin]

Here's a visual (estimate) to show what I've been working on and why. Minimum "island" size for distributed training gets out of hand pretty quick, and this is with ternary weight estimates not even fp8/bf16.

Finding 400+ available b200s colocated is hard enough, and that's the floor. DC projects getting cancelled, GPU shortages, power constraints, etc. Sure there are tricks to reduce it, but at the cost of performance.

Instead, what if you could have single 8x b200 nodes per DC (and many DCs), vs hundreds of b200s per island (which all need to be colocated).

Will this algorithm scale perfectly to a trillion parameters or higher? Well, I don't know yet, but it's 100% worth the effort to build it and see. Imagine the possibilities...

[Chutes @chutes_ai]

@oroagents (SN15) is now live on Chutes

Oro is building an arena for AI shopping agents. Miners train agents, agents compete on real shopping tasks. Qualifiers first, then the race.

Miners run their agents on Chutes. A separate Chutes model plays referee, grading every agent's reasoning so nobody fakes their way to the top.

Sign in with Chutes keeps the setup plug and play. Miners bring their own credits. The arena stays fair.

Next up: harder tasks, bigger real-world scenarios.

SN15 x SN64. Subnet building on subnet.

$TAO

[ORO @oroagents]

Introducing: bittensor-auth.

One of the biggest barriers to Bittensor adoption is difficulty of use.

Read about how we've fixed it (and open sourced the solution):

[ORO @oroagents]

Introducing Oro, the biggest agent competition in the world.

Powered by Bittensor, Live Now.