Openclaw Kimi K2 5 On Baseten Frontier Agent Performance With Oss

OpenClaw + Kimi K2.5 on Baseten = frontier agent performance with OSS Announcing our Series F . Learn more

AI engineering

OpenClaw + Kimi K2.5 on Baseten = frontier agent performance with OSS

OpenClaw + Kimi K2.5 on Baseten: frontier agent performance with open-source models. Outperforms Opus 4.5 and 8x cheaper. Install in 2min.

Authors

Alex Ker

Paras Stefanopoulos

Last updated February 4, 2026

Share

OpenClaw, previously ClawdBot and MoltBot, has taken over the internet as the first Jarvis-like agent that can actually get things done. If you’re not familiar with OpenClaw, think of it as  your personal assistant or teammate. OpenClaw runs on your local machine (or the cloud), and executes sub-agents with the ability to maintain memory and use tools. That means OpenClaw can do just about everything: write, execute scripts, browse online and use apps. You can even monitor its behavior from any chat of your choice, such as Telegram and Whatsapp. Why Kimi K2.5 & OpenClaw is a powerful pair: OpenClaw is an incredible everyday agent when paired with OSS inference, especially Kimi K2.5 on Baseten Model APIs. By pairing OpenClaw and Baseten’s Kimi K2.5 you get quality on par with closed source models and better performance all at a fraction of the cost. On agentic benchmarks, Kimi K2.5 outperforms Opus 4.5 . Kimi K2.5 costs $3/1M output tokens vs. Claude's $25/1M, that's 8x the savings. That can add up extremely fast for token-hungry applications like OpenClaw. ✕ Kimi K2.5 vs. Claude Opus 4.5 on agents and coding benchmarks ✕ Kimi K2.5 is 8x cheaper and 4.5x faster than Claude Opus Unleash the Claw (with Kimi), in only 2 minutes: If you’re interested in utilizing Baseten’s Kimi K2.5 Model API in OpenClaw, we’ve added a step by step guide below. All you need is a Baseten API key to get started ( here ). Add this to your environment or save it in a secure location before you dive in! Clone Baseten fork and launch OpenClaw Enter the following into your terminal: git clone  https://github.com/basetenlabs/openclaw-baseten.git cd openclaw-baseten pnpm install pnpm ui:build  # auto-installs UI deps on first run pnpm build pnpm openclaw onboard --install-daemon Follow the setup You should see this as the initial screen: ✕

Select QuickStart as onboarding mode ✕

Select Baseten from Model/auth provider and paste your Baseten API key generated before. ✕

After entering your API key details, you should keep the current model (as we recommend Kimi-K2.5). Alternatively, GLM-4.7 is a powerful agentic model. ✕

Continue through the rest of the steps, optionally setting up links to apps or skills. If you have a gateway previously running, it may interfere with this instance, so please select restart.

Hatch the bot in TUI ✕

You now should automatically be redirected to a webpage where you start interacting with OpenClaw. If you prefer the CLI, you can also chat with OpenClaw in the terminal. If you have set up Telegram or Whatsapp, you can control your agent through those apps! ✕

The bottom line: You no longer need to choose between performance, quality and cost. You can get access to higher throughput, lower latency, all while enjoying the smartest OSS agent in your pocket. You can power OpenClaw with any frontier open-source models like Kimi-K2.5, GLM-4.7, and GPT-OSS 120B on Baseten with this repo today.

Subscribe to our newsletter Stay up to date on model performance, inference infrastructure, and more. ‌

Explore Baseten today Start deploying Talk to an engineer

Related posts View all AI engineering

AI engineering How to run GLM-5.2 in any harness

Alex Ker

AI engineering Rolling deployments for zero-downtime model updates

Archit Mishra 2 others

AI engineering Cost-efficient, high-performance TTS with Qwen3-TTS

Ian Carrasco 1 other

Popular models GLM 5.2

Kimi K2.7 Code

DeepSeek V4

GPT OSS 120B

Whisper Large V3

NVIDIA Nemotron 3 Ultra

Explore all

Popular models GLM 5.2

Kimi K2.7 Code

DeepSeek V4

GPT OSS 120B

Whisper Large V3

NVIDIA Nemotron 3 Ultra

Explore all