Introduction to Open-Weight Models
OpenWeight models are getting near frontier quality
Tom W.
Scout A. TeamThree open-weight models released in recent months are strong enough to use on real coding work right now.
They're not "impressive for open source," they're just good models. If you haven't tried them yet, here's what to deploy first and how to launch them in about five minutes.
The Models
MiniMax M2.5 is a mixture-of-experts model from China that punches well above what most people expect from an open-weight release. It's strong on reasoning and coding tasks and handles long contexts without falling apart.
Kimi K2.5 comes from Moonshot AI with agentic use cases in mind. It holds up well on multistep tasks, tool use, and code generation. If you're building anything that involves an agent doing real work across multiple steps, pay attention to this one.
GLM-5 is the latest from Zhipu AI's GLM series. It's a capable general-purpose model that competes on coding, reasoning, and instruction following, and it can run locally.
All three show up near Opus 4.5 on public benchmarks, depending on the evaluation and harness. The gap between open and closed at the frontier is narrowing fast.
What Is Ollama Cloud?
Ollama started as a local model runtime, a clean way to pull and run open-weight models on your own machine. It's become the default tool for that workflow, and for good reason. The CLI is simple, the model library is comprehensive, and it's unobtrusive.
Ollama Cloud extends that with hosted model access. It has the same interface but with more computing power, running on Ollama's infrastructure instead of your laptop. That matters when you need stable, fast inference for longer coding sessions.
Create a free account at ollama.com for access to the model library immediately with no sales call and no waiting.
Getting Started
The fastest path to deploying any of these models with your existing coding workflow is Ollama's Launch feature.
Launch lets you configure a model for your favorite coding harness locally. Whether you're using Claude Code, Cursor, VS Code, Zed, or something else, Launch gives you a simple path that wires the model into the tool you already use. Pick the model, point your harness at it, and go.
Here's the short version of how to deploy Ollama Cloud:
- Go to ollama.com and create a free account.
- Pull one of the three models. From your terminal:
ollama pull minimax-m2.5,ollama pull kimi-k2.5,orollama pull glm-5. - Use Ollama's Launch feature to configure your coding harness to use that model locally.
- Start coding. The harness works the same way it always has, just with a different model.
That's it. There's no new interface to learn and no workflow to rewrite. You're only swapping the model underneath.
Which One To Try First
If you're doing agentic work or building anything with multistep tool use, start with Kimi K2.5. It was built for that use case, and it shows.
If you're primarily writing and reviewing code, MiniMax M2.5 and GLM-5 are both strong. Run both on your real workload for a few days, and see which one feels better. Benchmarks are a starting point, but daily tasks are the real test.
You don't need a migration plan to test these. Pick one model, wire it into your existing setup, and run your normal workflow this week.