Claude Code with Kimi K3.1 Pro: 43% cost cut
Swapping only the model behind Claude Code to Kimi K3.1 Pro yields a real 43% saving, with JEV resolving micro-forks in code before escalation.
THIS IS A PURE F*CKING GOLDMINE You can run Kimi K3.1 Pro inside Claude Code and pay half of what Opus costs. Same harness, same subagents, same tools. You only swap the model. Same session, two ways: 120k tokens cost $0.15 on Opus 5.5 and $0.09 on Kimi K3.1 Pro. here's the split: → Kimi K3.1 Pro runs the main agent, edits code and runs tests → search, edit and test subagents all run on the same model → 1M context loaded, so a big repo fits without constant re-reading → /status confirms the base url, the model and the auth before you trust it the part most people get wrong when they see the price list: they expect -50%. Opus is $4 in and $20 out per million. Kimi is $2 and $10, exactly half on paper. but a cache hit costs $0.20 on both, so the cached part of your session never gets cheaper. real saving lands at -43%, not -50%. the harness does not care which model is behind it. the same subagent files, the same tools, the same fork layer, only the weights change. one thing to know: this is not supported by Anthropic. your code goes to Moonshot, so decide that part yourself before you point the base url anywhere. same Claude Code. different model. half the bill. JEV engineering is the same move one layer down: 1,301 forks this run - which file, which tool, retry or stop - resolved in code at 0.90 sharpness, and only the split ones ever reach the model the full tree ▸ Claude Code runs as usual, only the base url and model change ▸ Kimi K3.1 Pro runs the main agent on a 1M context window ▸ search, edit and test subagents run on the same model ▸ /status verifies base url, model and auth before any real work ▸ cache hits cost the same on both sides, so the saving lands at -43% ▸ JEV resolves micro-forks in code and escalates only the split ones paste this into Claude Code ↓ "Set my Claude Code up to run on Kimi K3.1 Pro: 1. Point ANTHROPIC_BASE_URL at the Moonshot endpoint and set the model to kimi-k3.1-pro. 2. Keep my existing subagents and tools untouched, only the model changes. 3. Run /status first and confirm base url, model and auth before any real task. 4. Show me the per-million prices for input, output and cache hits side by side with my previous model. 5. Warn me explicitly about anything that leaves my code outside Anthropic. Show me every change as a diff first. No edits until I say go." ↳ http://code.claude.com/docs/en/model-config save it, this one pays for itself on the first session.
This post quotes another post: x.com/sprytixl/status/2107763347097534680