Ivan Neustroev

Synthetic.new is one of the less known inference subscription providers, but I've been using it for a while and I think it deserves more attention. What I like: - Very reliable inference in my experience, with minimal random failures or broken tool calls. - Weekly limits regenerate continuously, 2% at a time, instead of resetting all at once. I kind of prefer this system, it leads to healthier usage patterns. - It's particularly good for Kimi K3 and GLM 5.3 Flash right now. GLM 5.3 Flash is so cheap that my limits regenerate almost as fast as I spend it. - Pricing is flexible. No Standard/Pro/Ultra tiers. Instead it's based on $30 subscription packs, and you can stack from 1 to 5 depending on how much usage you need. The main downside: usage per dollar isn't the best you can find. Some subscription providers can give you considerably more. The problem is what those providers offer can also change dramatically at any moment. Ollama Cloud's $20 plan used to be plenty, then the limits got nerfed into the ground. Meanwhile Synthetic has stayed relatively consistent while adding better models over time. They also shipped Kimi K3 before Ollama did. So my take on Synthetic is pretty simple: - Reliable - Predictable - Transparent - Not necessarily the cheapest At this point, I value the first three quite a lot. If you got interested, I'll put my referral link below. If you subscribe through it, we both get $10 in subscription credit, 1/3 of a pack's price.