I’ve been so impressed with Kimi K2’s performance so far when it is working for me. I am currently locked out of my account after using it only for a couple of hours yesterday and about 5 minutes this morning. Am I out of quota for a full week? If so, this seems much less access than in my Claude $19/mo plan where it is every 5 hours it will reset.
Issue #2: Is there a 128 tool limit? I wonder if that is coming from Claude? I’ve never seen this in any of my projects before testing K2. It could be coming from VSCode’s Claude Code extension though.
“API Error: 403 {“error”:{“type”:“permission_error”,“message”:“resource_exhausted”},“type”:“error”} · Please run /login” within Claude Code.
”API Error: 400 {“error”:{“type”:“invalid_request_error”,“message”:“a max of 128 tools can be supported”},“type”:“error”}”
Apologize for the delayed response regarding your K2 experience. I’m happy to clarify both issues:
Regarding Quotas:
You’re not locked out for a full week. The K2 series models operate on per-minute rate limits (TPM/RPM), not weekly quotas. Your access automatically resets each minute based on your tier. You can check the specific limits for your plan here: https://platform.moonshot.ai/docs/pricing/limits.en-US
Regarding the 128 Tool Limit:
This limitation has been removed recently! The K2 series models now support unlimited tools, with the only constraint being the total prompt token size (the overall input tokens for the entire request, including messages). Please try again and let us know how it works for you.
If you continue experiencing access issues, please share your current usage tier and any error messages, and we’ll investigate further.
You’re absolutely right. Early LLMs struggled with too many choices, so after some trade-offs we temporarily set this 128-tool limit. However, with further training enhancements and the growing demands of Agentic AI, we’ve gained sufficient confidence to lift this restriction in our newer models like Kimi K2. The practical limit is now determined by context length rather than an arbitrary cap.
Please try again with the latest model—we’d love to see how it works for your use case.
In response to specialized scenarios like Claude Code, we’ve launched Kimi Code. You can find pricing plans (monthly or yearly) with dedicated token allocations on our membership page: Kimi - 会推理解析，能深度思考的AI助手
Important: To use Kimi Code, please read the documentation carefully and use the dedicated API key and endpoint from your Console, such as https://api.kimi.com/coding/ (not the standard endpoint). These specialized plans typically come with more generous usage limits.