The cheap open models are out there. Actually using them is the annoying part.

Short summary

A founder introduces Khidi, an OpenAI-compatible gateway that routes requests to cheap open-weight models like Kimi K2.6 at a fraction of flagship API costs. The post includes a pricing comparison showing ~10x savings for typical coding-agent workloads and benchmark data showing open models holding their own on SWE-Bench Pro. The gateway supports Cursor, LangChain, Claude Code, and other tools via a simple base-URL swap.

• •Khidi is an OpenAI-compatible gateway to cheap open-weight inference, claiming ~10x cost savings
• •Pricing comparison: $550/mo on OpenAI vs $45/mo via Khidi for typical coding-agent usage
• •Open models like Kimi K2.6 score competitively on SWE-Bench Pro and DeepSearchQA benchmarks
• •Founder disclosure: author runs khidi.ai, so content is partially self-promotional

Generated with AI, which can make mistakes.