Minos Opens Bittensor Subnet 107 Mac Beta Without Rewards

Minos opened a public macOS beta on November 7, 2026, for Bittensor Subnet 107, with on-chain subnet rewards inactive, according to the project’s official announcement and product pages. The downloadable Apple Silicon client requires macOS 14 or newer.

The release adds a prototype miner for Subnet 107’s next workload, known as evaluation authoring. Minos already runs Subnet 107 as a decentralized genomic variant-calling market, where miners identify DNA mutations in sequencing data and validators score accuracy against truth sets. The Mac app does not enter that live caller competition.

The economics remain explicit. The product page states: “public beta, subnet rewards are not active yet.” In the same morning’s official post, Minos thanked early testers and directed prospective users to its website for downloads.

What the client actually does

The product labels the process Research, Draft, and Recheck:

• Research: The Mac selects an assignment from Minos’s test queue, and the user’s own AI drafts a scientific brief from public literature.
• Draft: The Mac audits the draft locally. Source material and inference keys remain on the machine, with keys stored in macOS Keychain.
• Recheck: One signed evidence bundle leaves the Mac for each finished assignment through a signed upload URL.

The client also sends worker-status and activity events to the Minos platform API over HTTPS. It transmits the computer name during registration, adding an administrative identifier to the workflow.

Connected AI services include Codex through a ChatGPT account, Claude Code and Kimi Code through their command-line interfaces, and SayGM and Chutes through API keys. Users pay through their own subscriptions or credits. No work from this beta path is registered or submitted on-chain while rewards remain inactive.

Why the design uses Bittensor

Variant calling identifies which DNA bases differ from a reference genome. laboratories already use tools such as GATK, DeepVariant, FreeBayes, and BCFtools. Minos states that accuracy should be earned through open, scored rounds rather than self-reported inside closed pipelines.

Bittensor would provide the economic coordination layer through independent miners, scored work, and TAO-linked subnet emissions once a workload goes live. The present beta tests the surrounding client and evidence pipeline, not that incentive mechanism.

AI has a narrower role here than the genomics branding may suggest. The client does not ask a model to identify variants from a BAM file. It asks the model to write literature briefs that support evaluation authoring, after which local software checks and packages the evidence. That process does not establish clinical correctness or prove that a future variant call matches biology.

What the release proves—and does not

The November 7 post and company documentation establish a public client surface: a Mac app, a bring-your-own-AI path, local key storage, and an evidence-upload flow tied to Subnet 107’s next workload.

They do not establish the larger genomics thesis. Continuous, incentive-aligned benchmarking of variant callers, broader coverage of difficult genomic regions, and later phases involving custom callers, consensus AI, and production services remain roadmap claims.

Nor does the client prove that an AI brief is scientifically correct. It demonstrates that a machine produced a draft, local software executed a recheck path, and a signed evidence bundle reached Minos’s API. The work remains off-chain while rewards are inactive.

What does the beta change for miners and researchers?

For prospective users, it changes the operational requirements, not the economics. Participants must install the app, connect a supported model service, spend their own subscription or API credits, perform local processing, and submit evidence without receiving TAO-linked compensation.

For researchers, it adds a literature-and-evidence preparation lane for evaluation work. It does not yet show better genomic accuracy or establish a paid decentralized science market. The release tests whether people will install the app, connect model accounts, and ship signed bundles; it does not test whether Bittensor emissions can buy more accurate variant calling at scale.

The experiment also sits alongside other decentralized-AI products that recast the role of a stake or client, including 0G’s path from staked assets to AI compute credits. In Minos’s case, the immediate test is whether signed evidence bundles become on-chain scored jobs and whether those scores correspond to measurable improvements in real variant calls. The next checkpoint is structural rather than promotional, with subnet rewards still disabled during the public-beta phase.