Models / Use cases
Best open source models, open-weight LLMs you can run anywhere
Open weights mean you can leave any time, self-host later, swap providers, or keep calling them through one metered endpoint today. These are the open models that hold up in production.
Ranked
The picks
DeepSeek V4 Pro
DeepSeek
The strongest general open-weights model per dollar.
Context
1M tokens
Max output
393K tokens
Tools
Yes
View DeepSeek V4 Pro pricing and playground →
Kimi K3
The open pick for agentic tool-calling workloads.
GLM-5.2
The open coding-and-reasoning challenger, a real alternative to proprietary coding models.
GLM-5.3-Flash
The cheapest open model here, a million-token window at a price that makes bulk processing trivial.
DeepSeek V4.1 Flash
The V4 family's volume tier, open weights at commodity per-token pricing.
Replaces DeepSeek V4 Flash, retired by DeepSeek on 2026-09-10.
Retired
Previously recommended here
The V4 family's volume tier, open weights at commodity per-token pricing.
Use DeepSeek V4.1 Flash instead, same endpoint, same key, one line changes.
FAQ
Frequently asked questions
Why call an open model through an API instead of self-hosting?
Self-hosting pays off at sustained high volume. Below that, per-token API pricing with zero ops overhead is cheaper, and because the weights are open, the exit door stays open too.
What does open weights actually let me do?
Download the model and run it on your own hardware, indefinitely, with no dependency on the provider staying in business or keeping the model available. That is a real hedge, and the reason these models are worth preferring where quality is comparable.
Are open models good enough for production?
For most workloads, yes. The gap against the frontier tier is narrowest on coding, extraction and structured output, and widest on long multi-step reasoning. Compare them on your own prompts before assuming either way.
Can I use the output commercially?
Each model page carries the provider's own license summary. Open weights and permissive output licensing usually travel together here, but the terms are the provider's, so the page states them rather than generalising.
What happens if a provider retires an open model?
We keep its page, say plainly that it is retired, and link the replacement, which is exactly what happened to DeepSeek V4 Flash. With open weights you also keep the option of running the retired model yourself, which you do not get with a proprietary one.
One key. Every model. Exact prices.
Route every pick on this page through one key. Free starter credits included, no subscription required.