Minos Opens Bittensor Subnet 107 Mac Beta Without Rewards

Bittensor Subnet 107’s new macOS beta offers no rewards and makes no on-chain submissions. Minos announced the public beta at 11:50 p.m. ET on Friday, Nov. 6, or 4:50 a.m. UTC on Nov. 7, according to the company.

The prototype opens access to a proposed evaluation-authoring workload. It supports only Apple Silicon Macs running macOS 14 or later. Participants must bring their own AI subscription or credits, shifting those service costs to testers while the beta generates no subnet compensation.

The public beta gives users access to a prototype for a proposed workload, but it does not yet extend Subnet 107’s on-chain work, reward distribution or validation duties.

How the test runs

A user receives a scientific-literature assignment through Minos’ platform test queue. Minos said in its announcement: “Assignments go to Minos’ platform test queue.” The wording locates the workload inside Minos’ test infrastructure, not Subnet 107’s on-chain submission flow.

The external AI service drafts a scientific brief from public literature. The Mac audits that draft, then uploads an evidence bundle through a signed URL. Listed connections include Codex, Claude Code, Kimi Code, SayGM and Chutes, while users remain responsible for supplying the subscription or credits each service requires.

The sequence combines public scientific literature, an external AI draft, a Mac-side audit and a signed-URL evidence upload, all without requiring the participant to register or submit work on-chain.

Minos also states that “participants receive no subnet rewards.” The beta therefore tests queue handling, drafting, auditing and evidence transfer. It does not connect those steps to the subnet’s established validation process or reward path.

The established workload

Subnet 107’s established work centers on genomic variant calling. Miners run variant-calling tools on genomic data, and validators score the results against truth data. Minos presents evaluation authoring as a proposed next workload.

Evaluation authoring would sit beside a genomic system in which miners process variant data and validators compare the resulting calls against truth data for scoring.

The proposed workload uses a different output. It asks an AI service to produce a literature-based brief and supporting evidence, rather than a genomic call. The beta’s briefs and uploads do not show whether the briefs are scientifically correct or whether evaluation authoring improves variant-calling results.

What does the beta change for Subnet 107?

For participants, the immediate option is a way to test the proposed workflow on supported hardware, rather than a way to earn Subnet 107 rewards or build an on-chain submission record.

The answer is narrow. The beta creates an off-chain path for testing a workload that could eventually join Subnet 107’s existing task set. It changes neither the reward terms nor the validator role during the test. The announcement also sets no date for moving the workload into live validation.

Testing the client also does not prove the method’s value. The beta tests the queue-to-evidence workflow, but its outputs do not establish scientific quality or gains in variant-calling performance. The announcement provides no result showing that the briefs are accurate or that the resulting evaluations would improve variant calling.

What the announcement does establish is operational: the required hardware and model access are defined, the assignment path runs through Minos, and the output includes an audited draft plus an uploaded evidence bundle. The test avoids the on-chain machinery tied to established subnet work.

Until Minos runs evaluation authoring through Subnet 107’s on-chain validation and reward path, the practical effect remains limited to workflow testing. No deadline is stated for that market-structure change.