Bittensor Ecosystem Highlights :: July 20–26, 2026
[ @chutes_ai - Subnet 64 ] Chutes pre-trained a 20B model for under $10 an hour without a dedicated cluster.
Parallax coordinated rented GPUs across two continents and came close to the quality of a model trained on four high-end datacenter GPUs. It shows that affordable GPUs in different locations can train large AI models, a job that normally requires a massive datacenter cluster.
[ @TargonCompute - Subnet 4 ] TargonOS is now live, bringing confidential AI compute to Bittensor.
Developers can now rent a private machine, install what they need, and keep their data hidden from the hardware provider. This gives Bittensor a decentralized cloud with privacy enforced directly by @Intel hardware.
[ @engyai - Subnet 53 ] Engy is now live on Bittensor, serving frontier open models with verified inference.
Its cryptographic proofs let users check that the model they chose produced each answer. Engy combines low-cost compute with SOTA cluster management to deliver some of the cheapest frontier-model inference on the market through Bittensor.
[ @bitmind - Subnet 34 ] BitMind published a new paper testing its deepfake detector across 19 public benchmarks.
It matched the best commercial image detector, beat the best commercial video detector, and outperformed leading published models on several major benchmarks. The system comes from Bittensor’s SN34 competition, where miners continuously generate harder fakes and build better detectors as deepfake technology evolves.
Miners will score the surface, corners, edges, centering and overall condition against grades set by professionals. The challenge puts Bittensor vision models to work on a real commercial task where tiny defects can decide a card’s final grade.
[ @actualinc - Subnet 95 ] Actual Computer joined the NVIDIA Inception Program. The program can give the team access to NVIDIA’s technical resources, experts and go-to-market support as it expands fast local inference across consumer GPUs.
70·B+Long
n
news7/26news
Bittensor Ecosystem Highlights :: July 20–26, 2026
[ @chutes_ai - Subnet 64 ]
Chutes pre-trained a 20B model for under $10 an hour without a dedicated cluster.
Parallax coordinated rented GPUs across two continents and came close to the quality of a model trained on four high-end datacenter GPUs.
It shows that affordable GPUs in different locations can train large AI models, a job that normally requires a massive datacenter cluster.
[ @TargonCompute - Subnet 4 ]
TargonOS is now live, bringing confidential AI compute to Bittensor.
Developers can now rent a private machine, install what they need, and keep their data hidden from the hardware provider.
This gives Bittensor a decentralized cloud with privacy enforced directly by @Intel hardware.
[ @engyai - Subnet 53 ]
Engy is now live on Bittensor, serving frontier open models with verified inference.
Its cryptographic proofs let users check that the model they chose produced each answer.
Engy combines low-cost compute with SOTA cluster management to deliver some of the cheapest frontier-model inference on the market through Bittensor.
[ @bitmind - Subnet 34 ]
BitMind published a new paper testing its deepfake detector across 19 public benchmarks.
It matched the best commercial image detector, beat the best commercial video detector, and outperformed leading published models on several major benchmarks.
The system comes from Bittensor’s SN34 competition, where miners continuously generate harder fakes and build better detectors as deepfake technology evolves.
[ @webuildscore - Subnet 44 ]
Score is launching a new SN44 challenge where AI models grade Pokémon and other trading cards from photos.
Miners will score the surface, corners, edges, centering and overall condition against grades set by professionals.
The challenge puts Bittensor vision models to work on a real commercial task where tiny defects can decide a card’s final grade.
[ @actualinc - Subnet 95 ]
Actual Computer joined the NVIDIA Inception Program.
The program can give the team access to NVIDIA’s technical resources, experts and go-to-market support as it expands fast local inference across consumer GPUs.