How to Run OpenClaw with Open-Source Models

2026-04-22 · Source: Towards Data Science · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

Anthropic recently banned using Claude Code subscriptions for OpenClaw, forcing users to rely on the more expensive Claude Opus 4.6 API, priced at $5 (input) and $25 (output) per million tokens. This prompted a search for cheaper alternatives, with OpenAI's GPT-5.4 proving "lazy." The article explores Chinese open-source LLMs like Kimi-K2.5, GLM-5.1, and MiniMax-M2.7 as viable options. Kimi-K2.5, available via OpenRouter for approximately 0.6/3 USD per million tokens (about 1/10th the cost of Claude Opus 4.6), is highlighted for its performance. The setup involves obtaining an API key from OpenRouter and configuring OpenClaw, ensuring all previous Anthropic references are removed to avoid OAuth issues. While Kimi-K2.5 is slightly slower than Claude Opus 4.6, its overall performance is comparable, making it a strong cost-effective competitor.

Key takeaway

For AI Engineers and ML Directors seeking to reduce operational costs for OpenClaw assistants, migrating from Claude Opus 4.6 to open-source alternatives like Kimi-K2.5 offers substantial savings. While Kimi-K2.5 might exhibit slower response times for simple queries, its overall performance is competitive, making it a viable choice. Be mindful of GDPR compliance if using Chinese models via API for sensitive data, or consider self-hosting for full control.

Key insights

Cost-effective open-source LLMs like Kimi-K2.5 can power OpenClaw effectively, despite minor performance trade-offs.

Principles

• LLM cost significantly impacts operational viability.
• OpenClaw is largely language model-agnostic.
• Remove all prior model references during LLM migration.

Method

To integrate Kimi-K2.5 with OpenClaw, obtain an API key from OpenRouter, configure OpenClaw via Claude Code, and crucially, remove all prior Anthropic API keys and environment variable references to prevent OAuth issues.

In practice

• Consider Kimi-K2.5 for OpenClaw to reduce API costs.
• Use OpenRouter for flexible access to multiple LLMs.
• Implement cron jobs for continuous model learning.

Topics

• OpenClaw
• Kimi-K2.5
• Claude Opus 4.6
• Open-Source LLMs
• API Pricing

Best for: AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• How to Run OpenClaw with Open-Source Models — OpenClaw can be powered by alternative LLMs like Kimi-K2.5 following Claude's subscription ban.
• Open model Kimi K2.7 Code undercuts GPT-5.5 and Claude by up to 12x on price per token — Kimi K2.7 Code offers a cost-effective, agent-optimized open-source model for complex coding, despite trailing top benchmarks.
• Kimi K2.6 with OpenCode & OpenRouter | Agentic RAG with LangChain, LangGraph & NextJS | 🔴 Live — Kimi K 2.6 demonstrates strong coding capabilities within an agentic environment, but provider choice significantly impacts performance and cost.
• Kimi K2.6: BEST Opensource AI Model That Beats Opus 4.6 and Gemini 3.1 Pro (Fully Tested) — Kimi K2.6 is a cost-efficient, open-source coding model with "state-of-the-art" long-horizon execution and multi-agent capabilities.
• How I Choose Between GPT, Claude, Gemini and Open Source Models for Every Task — Optimizing LLM usage requires a decision framework matching tasks to models for cost and quality.
• Run a Local LLM with OpenClaw on Your Mac Mini — Running a local, quantized LLM on a Mac Mini with llama.cpp significantly reduces OpenClaw API costs while maintaining performance for common agent tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.