> ## Content Index
> Fetch the complete content index at: https://www.tao.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# TPN Labs Releases First Production Model From Bittensor Subnet 65
- URL: https://www.tao.media/tpn-labs-releases-first-production-model-from-bittensor-subnet-65/
- Published: 2026-10-12T18:33:51.000Z
- Updated: 2026-10-12T18:33:52.000Z
- Description: The first TPN v0.2 winner compresses Ministral-3-8B by about 40% while retaining its full 32,000-token context window.
- Author: Tristan Hillerich
- Tags: TPN Labs, Subnets, AI, News

[True Performance Network](https://x.com/TPN%5FLabs/status/2109627952514175330?ref=tao.media), the team behind Bittensor Subnet 65, says its model-compression market has produced its first release-ready model.

The model, TPN005, came out of the subnet's fifth competition and is derived from Ministral-3-8B. TPN Labs said the release cuts the base model's size by about 40% while preserving the full 32,000-token context window, making it the first production artifact from a subnet designed to reward miners for shrinking open models without materially degrading their usefulness. The release also marks the first winner scored entirely under Fluid Bench, the private evaluation system introduced with TPN v0.2.

## TPN v0.2 Also Changed the Mining Mechanics

Under TPN v0.2, miners run the benchmark suite themselves through TPN Bench and submit run IDs with their models. Validators then check those records against TPN's backend instead of rerunning every benchmark from scratch.

That design is meant to reduce validator workload while preserving an auditable submission trail. It also makes the quality of the benchmark and backend verification process central to the subnet's credibility, since compressed-model competitions depend on reliable comparisons between submissions.

[SubnetRadar's summary](https://subnetradar.com/subnet-news/65/2026-10-06?ref=tao.media) said the v0.2 rollout removed a 0.3 TAO collateral requirement, set the registration burn at 0.2 TAO, and changed rewards so only the top three finishers are paid, with a 60%, 25%, and 15% split. That steeper reward curve concentrates rewards on the strongest submissions instead of spreading emissions across the field.