Subnet 97: Albedo
Albedo is Bittensor Subnet 97. Its on-chain identity describes the subnet as “alchemical intelligence” and lists unarbos/albedo as the subnet’s GitHub repository.
What Albedo Provides
The Albedo mining guide describes the subnet as a “king-of-the-hill” competition for language models. At any time, one model holds the title of reigning king. Miners submit challenger models that are matched head-to-head against the king on a fixed set of coding tasks drawn from a public dataset, and an ensemble of independent large language models judges which side performs better. A challenger that beats the king takes the crown and becomes the model that others must then beat.
The Albedo website index frames the subnet as a trajectory-distillation network: matches publish turn-by-turn transcripts, not just final scores. Those transcripts are openly available, so that researchers can study the winning behavior or use it to train new models.
Miner and Validator Roles
Miners compete by training and submitting challenger models that aim to outperform the current king. The mining guide describes a challenger as taking the crown when the judges favor it on at least one scoring dimension without losing on any other, so improvements can accumulate across successive champions.
Validators organize the competition: they detect new challengers, run each match against the reigning king, and record the outcome on-chain by setting weights. Rewards follow a “king chain” in which emissions are shared between the current champion and a handful of recent former champions that remain on the subnet. Operators should treat the repository as the source of truth for current participant guidance.
King Chain and Evaluation Context
The public mining guide and website index describe Albedo as a rolling king-chain competition. The active king is the model a challenger must defeat, and a winning challenger can become the new reference model for later duels. The website index also describes a five-slot king chain, so recent successful models can remain part of the reward path instead of disappearing as soon as a new winner is crowned.
A submitted model moves through several source-described stages before it can receive that role. The mining guide says miners publish a model artifact and on-chain reveal, validators validate the artifact, and the evaluation system compares eligible challengers against the reigning king. The website index describes the later coronation and weight-setting stages that turn a winning duel into subnet emissions.
The SWE-ZERO dataset source clarifies the benchmark setting. The dataset contains large-scale agentic coding trajectories from software-engineering pull-request snapshots. In Albedo, those trajectories provide coding tasks where the challenger and king can be compared on generated behavior, while the website index describes transcript artifacts that preserve the evaluation record.
Together, those sources show why the subnet is not only a model registry. Albedo links a miner’s published model, validation status, judged king duel, transcript output, coronation, and weight publication into one competition cycle. The reward target is the model that can become and remain king under that cycle.
References: Albedo mining guide, Albedo website index, SWE-ZERO dataset
On-Chain Identity
Live SN97 data is available on TaoStats. The live Finney identity for netuid 97 reports the subnet name as Albedo, the description “Alchemical intelligence,” and the GitHub repository as unarbos/albedo.
Relationship to Yuma Consensus
Subnet 97 uses Yuma Consensus to convert the king-chain competition weight vectors that validators submit into the emission shares distributed to miners and validators within the subnet each tempo. The Yuma Consensus documentation describes how validator weight submissions are aggregated into consensus weights for each miner registered on the subnet.
In Albedo’s context, validators run head-to-head king-of-the-hill duels between challenger models and the reigning king, distributing weight among the current champion and recent former champions in a king chain based on duel outcomes. The Emission documentation describes how those consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Albedo (SN97), that sequence changes how readers should interpret king-of-the-hill language model competition examples and duel-scoring outcomes.
In localnet, Albedo-compatible miners and validators can be developed and tested in an isolated environment. Localnet model duel results and emission outcomes do not represent production subnet performance.
On testnet, Albedo-compatible challenger model submissions can be exercised in a shared, non-production network. Testnet king-chain results and validator weights are separate from mainnet subnet state.
On mainnet, Albedo (SN97) is the live production subnet where miners submit challenger language models and validators run head-to-head king-of-the-hill duels to determine real Bittensor emissions. The Albedo repository describes the mechanism that applies on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A model duel result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Netuid 97 Identifies the Subnet On-Chain
Bittensor assigns every subnet a unique numeric identifier called a netuid, and Subnet 97 is the subnet registered at netuid 97 (Glossary: Netuid). The Understanding Subnets reference explains that each subnet runs its own incentive mechanism while sharing the same underlying Subtensor chain, so the netuid is the stable handle that distinguishes Albedo from every other subnet.
For a reader, this means “Subnet 97” and “netuid 97” refer to the same on-chain slot. A claim about Albedo should be tied to that netuid rather than to the registered name alone, because the name field can be changed on-chain while the netuid stays fixed.
Reader Boundary
Subnet 97 Albedo should not be read as generic Bittensor subnet documentation, an official ranking of commercial language models, or proof that the reigning champion is the best available model. It names one king-of-the-hill competition where miners submit challenger language models judged head-to-head against a reigning champion, with full match transcripts published on netuid 97 (Understanding Subnets, Glossary: Netuid).