Open-source model roundup: June 2026
Where the open-weight models stand this month, and what's worth trying — verify specifics.
Updated 21 Jun 2026 fast-moving — check a current source
Archived June 2026 snapshot, superseded by the July roundup. Kept for the record — the open-weight scene moves too fast for a month-old page to be useful.
This is an archived snapshot from June 2026. For the current picture, see July’s roundup.
The open-weight scene moves faster than anything else in AI, so a dated roundup goes stale quickly by design. The aim here is to track the direction and point you at things worth evaluating yourself.
What’s notable this month
- Small models punching up. Compact open models keep closing the gap with hosted frontier systems on everyday tasks — which makes running locally viable for more people.
- Better tooling. The runners that hide the setup complexity continue to improve, lowering the bar to a first local model.
- Retrieval gets easier. More turn-key options for embeddings and RAG over your own documents, locally.
How to evaluate a new release
Don’t trust a leaderboard alone. Pull a candidate, run it on your actual tasks, and weigh quality against the hardware it needs — and remember the energy trade-offs of running it yourself.
A model topping a benchmark this week tells you little about whether it’s right for your work. The only test that counts is your own use case on your own hardware.
Sources
Everything above was checked against these on 21 Jun 2026. Providers change things without notice — if a detail matters to a decision, follow the link.