Been using the free @opencode @nvidia agentic coding setup for a few days. Quick impressions: - GLM 5.1 works the best, but gets rate-limited very quickly - Minimax M2.5 is usable but needs constant supervision - Kimi K2.6 works but keeps getting interrupted mid-way - DeepSeek V4 Pro barely works (keeps thinking forever) - Couldn't get Minimax M3 to load at all -- although available on nvidia build but can't pull it into opencode Overall it's "fine" for being free. Anyone else testing free agent setups? Which models are actually stable for you?

OpenCode(無料)＋ NVidiaのNIM（無料のAPIキー） がおすすめ。 Kimi K2.6とか、無料で使える。

The OpenCode desktop app is unable to save/restore previous session histories when closed and relaunched. Looks like a bug. Hoping this can be fixed in the next update.

opencode should work, on the screenshot it showed you a better-sqlite3 missing error, i need to check why it would do it, but i can probably fix that easily. can add factory while at it

As long as only the score gets shared (: also include factory and opencode

Mais uma empresa construindo sua infraestrutura para codigo. Nada melhor que a propria empresa para entender como utilizar seu modelo proprietario do jeito mais eficaz. Espero que chegue logo o GLM 5.2 no @opencode para eu poder testar.

I think you're probably right. Means people like me have to work faster. Back to opencode dungeon

@vercel @opencode 🫶

Coming in OpenCode Desktop v2 10x faster performance (measured) I’m not kidding! I used a 30x cpu slowdown for my baseline at 4fps on an already weak machine, getting to approximately 45fps. - Migrated to tanstack virtual - Removed flicker, scroll (unintentional) jumping and lag while scrolling - Updated to the latest Pierre diffs version and shiki versions for free gains - Offloaded Shiki to a worker thread (off UI thread!!) - A number of performance tricks based on the hot path append only assumptions - Generic performance trace cpu time optimisations

Check out Omnigent, an open source harness that lets you use all the existing code harnesses (Claude Code, Codex, OpenCode, pi), collaborate and share sessions in many modalities (e.g. Slack/Teams, cli, webui), while having a fine grained security model that really tightens the control on what agents can do/not do. github.com/omnigent-ai/omnig…

Btw, I purchased Opencode subscription-it’s 5$ for the first month. And you can try the list of models through api

Omlx by far is the best inferencer for my M1 Max. It works seamlessly with @NousResearch's Hermes, @OpenWebUI, and @opencode, and can easily work with increasing context windows with models north of 30B parameters. Infinitely better than using LM Studio. omlx.ai/

Z.ai lite plan, Minimax.io lowest plan, opencode go and some openrouter usage.

It's for single API requests, not for agent workloads. You can't use this to power OpenCode or something of that nature. At most you can add a "consult an expert" one-shot MCP call.

Last few builds of OpenCode have been totally unstable for me and unusable. Hopefully the release tomorrow resolves it

Have not tried opencode or pi. My workloads aren't all that demanding, I'm just automating my itops work, systems engineer stuff. I'm also on windows, which has worked fine. I'm not aware what advantages pi or opencode may present, but I have been following the discourse loosely

Después de gemini, opencode toca probar Claude caude para probar cuando aportan los modelos frontera sobre todo en vibecodear ideas de principio a fin, primera suscripción grande que pruebo a ver qué tal va

Damn whatever router or provider OpenCode Go is using for inference, it must be slammed right now lol.