Chutes.ai
Chutes provides plans beginning at $3 as effectively. With Chutes, you could have entry to many extra fashions. Their efficiency is underwhelming. I’d suggest Chutes over Z.ai simply because you should utilize extra fashions, however that’s the one cause. Their privateness coverage is ambiguous. I had points signing up; mainly, authenticating with Google or GitHub isn’t working proper now. Primarily based on a advice from one other person on their Discord, I used a VPN and created an account with out utilizing Google or GitHub, and it labored.
Artificial
Artificial provides a $20-per-month plan that features GLM-4.6, Qwen3-480B, Kimi K2, DeepSeek-V3.1, and plenty of different fashions. Artificial is comparatively new, and so they’ve been tremendous open and interesting on Discord. They’ve a really clear privateness coverage: they don’t retailer your prompts or completions past 14 days with out your consent. Notice that there are points akin to software calls hanging (aka buffering) for a very long time and, often, leaking native software calls or software calls into the message stream slightly than correct OpenAI-compatible calls. You should use a CLI like LLxprt Code or Code Pet, or an IDE like Roo Code, that helps OpenAI endpoints. Artificial can be growing their very own coding agent referred to as Octofriend.
Cerebras
Cerebras plans begin at $50. I ripped them a brand new one in my final column, however they’ve tremendously improved their total TPM (tokens per minute) and request caps. Cerebras is now the quickest supplier bar none. Their insurance policies explicitly notice they don’t use your prompts. Cerebras is the quickest Qwen3-Coder-480B you possibly can entry. Whereas the Qwen3 mannequin helps 256K context, the Cerebras model is restricted to 131K. I’ve discovered that usable however difficult. Nevertheless, I nonetheless have a subscription in my software equipment as a result of it’s the quickest entry to probably the greatest open-weight fashions round. Cerebras has a free tier, however it’s barely sufficient for connection testing.
