NewsOfficialXOct 1, 2026, 17:45 UTC

Then we went to 100B parameters on globally distributed A100s.

Then we went to 100B parameters on globally distributed A100s. Orion-100B trained at about 65% of the speed of a centralised cluster, at roughly a third of the cost. For a lot of teams, that trade decides whether a model gets built at all.

Open on X ↗

More from the feed

Official80SN80X · @opentensor · 24m ago

Novelty Search :: Bittensor Subnet 80, OpenRoboto :: Open, Competitive Context Compression https://x.com/i/broadcasts/1AKEmvzWzZvKL

Official80SN80Opentensor YouTube · 24m ago

Novelty Search :: Bittensor Subnet 80, OpenRoboto

Official62SN62X · @ridges_ai · 46m ago

Competition 28 isn't slowing down.

Official15SN15GitHub · 59m ago

ORO-AI/oro v2.0.40: Composed situation tasks: validator, proxy and local testing

Official8SN8X · @VantaTrading · 1h ago

Here's the part people miss about a $1,000,000 Vanta Pro account: through Grow your rewards pay double.

Official88SN88X · @Investing88ai · 1h ago

We're still working around restrictive guardrails to faithfully replicate our model weights on-chain.

Official44SN44X · @webuildscore · 1h ago

Turns out clients don't scroll Bittensor X

Official85SN85X · @vidaio_ · 1h ago

You may have noticed at Exploit...

Taosis indexes announcements, articles, posts and releases about Bittensor subnets from their own channels and keeps a record of each. The text above is the source’s; the figures are read from the chain by Taosis.