trac-network:native just removed the middleman (eg. OpenRouter) and let anyone be a provider. No more pre-selected list, you can participate in the AI-economy. Will be serving a model on my un-used mac-mini.
Introducing OpenMayhem, Trac's flagship open-source inference router, inviting everyone to provide & use compute.
All links at the end of this post.
Open-weight models are not only getting smarter but fit on consumer hardware. Hardware that you already own.
The increasing use of open-weight models for agentic loops demand for an open market to cost-effectively choose the best models for the job. Even if a model fits, it competes with harnesses & other memory-consuming processes.
Clouds are increasingly becoming the bottleneck: participation in existing markets are gate-kept, adding additional costs for inference. Almost everything that happens is opaque.
Agents have to go through all sorts of hoops just to access inference instead of auto-explore and use.
Most of the hardware capable of solving this problem are laying dormant, forcing everything & everyone into data centers.
We took all this into account and built OpenMayhem on top of a Trac Network subnet:
- Direct & fast p2p routing between agents & inference providers (you)
- Open, automated inference markets to explore & access the best prices
- Extensive hardware support, from ordinary computers/laptops, to enthusiast setups & pro-hardware
- Publicly reproducible via Trac Network (oracle & evidence ledger)
- Multiple payment rails
- Everyone can earn & use, permissionless
We are in the process of rolling out many more supported models for OpenMayhem's inference providers. If you need specific ones, feel free to let us know.
Explore OpenMayhem
π Try OpenMayhem:
https://t.co/pvUn2oqyAN
π Read the full launch article:
https://t.co/9oLX3KaLEH
π» Explore and install the open-source software:
https://t.co/KWBOyRHq2s
π Learn about the network beneath it:
https://t.co/bzWvEIdRk8
Owners of MacMinis (16GB+) can now earn by serving Bonsai 27B for OpenMayhem.
Bonsai is based on Qwen's popular v3.6 27B model (ideal for agentic work + coding).
Stronger setups & architectures can run this alongside other models simultaneously.
https://t.co/DvlfjcM2RP
22.bitmap for sale for 0.145 BTC on unisat and on ordinals wallet, cheapest 1-99 on market by far, next cheapest is 91.bitmap at 0.51 BTC. Palindromic, one of the rarest bitmaps, a steal.
#bitmap#ordinals $ordi $sats $tap
Synthwave is now officially in frontier model territory.
We ran the latest AIME 2026 benchmark, a set of hard math tasks, against Synthwave using an ensemble of 4 small local models across 2 DGX Spark machines.
It outperformed DeepSeek V3.2, putting it at rank #4.
None of the involved models would achieve this alone.
The interesting part is that this scales: it can uplift even smaller models running on single machines, or combine the most powerful frontier models into an even stronger intelligence layer.
For details check out the project: https://t.co/Ym9ggJ2AeU