Open-weight 27B hits 38% on Terminal-Bench 2.0 (Opus 4.1 hit 38% in Aug 2025)

> today's best runnable-offline model is roughly 6–8 months behind today's frontier.

But it doesn't matter because frontier models were extremely good 8 months ago and we were doing real work with them. Now we have more capable open-source agents like pi and OpenCode which work well with these models.

More importantly, offline models is the best choice for privacy, on-device inference and no token/cost anxiety.

	▲	swrrt 3 hours ago \| parent \| next [-]
		Yep, offline mode is useful for edge devices also. I am considering deploying a extremely small model on steam deck actually.
	▲	ubermon 6 hours ago \| parent \| prev [-]
		Totally agree! I think we are very early on discovering the full potential of local models

▲

merkleforest 7 hours ago | parent | prev | next [-]

> 2. How local use feels in practice

Do we have stats on how does the models do on Mac M-series chips?

	▲	ubermon 6 hours ago \| parent [-]
		Not yet, will conduct a more comprehensive one later

▲

timothyshen123 8 hours ago | parent | prev | next [-]

Interesting find on this. Thanks for sharing

	▲	ubermon 6 hours ago \| parent [-]
		Thank you! I think there is a lot to dive deep later with different hardware, inference engine, prompt/harness setup etc.

▲

debpack 8 hours ago | parent | prev | next [-]

this is super sick man

	▲	ubermon 6 hours ago \| parent [-]
		Thanks!

▲

ubermon 9 hours ago | parent | prev [-]

[dead]