Today I discovered something pretty cool... OpenRouter has free models! And not just baby 2b models, 550b MoE from Nvidia (Nemotron 3 Ultra), along with some really cool ones I discovered along the way, like HY3 by TenCent and North Mini Code by Cohere.
This is a game changer... We all pay subscriptions, or API fees, I built V.E.L.O.C.I.T.Y. IDE to just get a cheaper alternative, using Kimi K2.7. But then I discovered, DeepSeek is shockingly cheap. You get a REALLY competent coding model for sub $1/M tokens in and out... But this... This is the real victory. V.E.L.O.C.I.T.Y. IDE still needs some polishing, but connected with a Cloudflare API key, or OpenRouter API key, it gives you access to any model you want! And that includes free ones.
I'll post an update once it's at a state where it's 'nice' to use, so you all can give it a try.
Top comments (8)
You can try OpenCode and its own model (Big Pickle—a rebranded GLM 4.6) or one of its free Zen models, like DeepSeek v4 Flash. It does 99% of what Copilot does using Haiku 4.5, and you don't even have to open your wallet.
Not gonna lie — you just found this out today?🤣
🫠 I did. I knew some platforms had free (but limited) models, practically all of them do, but the bit that's insane to me, is that it's free for such massive models? 550b MoE is a pretty capable model, just openrouter uses 10k requests a day cap... Good thing I built a multi-account balancer 😅 So between OpenRouter and Cloudflare with 20ish API keys each, I think there's a good chance I can get a decent amount of work done in a day? Foundry has been running off APIs, so it's been quite pricey when I push bigger projects, but this little hack is the perfect playground to test out the orchestrator layering for more complex tasks.
Passing on the spirit of freeloading, hahaha 🤣
With token prices where they're at now... Most definitely 😂 I mean look at what Fable charges!? Vs Deepseek costs less than a dollar a mil? Same size model, same level of intelligence, you just need a better harness and gating to get it to outperform Anthropic and literally get 50x the output:cost. Or you know... Free is apparently an option?
Absolutely — that money's better spent on a bucket of KFC anyway. 🍗
Whole stomachly agreed!
They say they have free but notice what they route you to and how quickly it degrades.