ClawKit LogoClawKit

OpenClaw itself is free — no subscription, no licence fee.

You pay only for the LLM API calls your agent makes. Use Gemini Flash or local Ollama to run at $0/month, or pick a paid model for better performance. ClawKit tools (Config Wizard, Doctor) are also free.

OpenClaw Pricing — LLM API Cost Comparison 2026

Updated February 2026 · Prices from official APIs

OpenClaw agents consume tokens on every step — tool calls, reasoning, retries. A 100-step task can cost anywhere from $0.00 (free models) to $5.40 (premium). This guide helps you pick the right model for your budget.

Free Models — Zero Cost, No Credit Card

You can run OpenClaw agents completely free. These providers offer generous free tiers that are enough for personal use and learning.

Google AI Studio

Gemini 2.5 Flash/Lite, Gemini 2.5 Pro

1,500 req/dayNo card

OpenRouter

30+ free models (DeepSeek, Llama, Gemma...)

200 req/dayNo card

Groq

Llama 3.3 70B, DeepSeek R1, Gemma 2

1,000 req/dayNo card

Ollama

Any open-weight model (Llama, Mistral, Hermes...)

Unlimited (local)No card

Mistral

Mistral Large, Mistral 8B

1 req/sec, 500K tok/minNo card

Cloudflare Workers AI

Llama 3.1 8B, Mistral 7B

10K neurons/dayNo card

New to OpenClaw? Start Here

Use Gemini 2.5 Flash (free tier) from Google AI Studio. No credit card, 1,500 requests/day, excellent tool calling, and a massive 1M token context window. Switch to DeepSeek ($0.04/100 steps) when you need more capacity.

Generate Config with Gemini

Full Model Pricing Table

Sorted by cost. "Agent Cost (100 steps)" assumes ~600 input tokens + ~400 output tokens per step — a typical OpenClaw browser automation task.

ModelProviderInput $/1MOutput $/1M100 StepsContext
Ollama (Local)100% Free
Your HardwareFreeFree$0.00Model-dependent
Gemini 2.5 FlashFree TierHOT
Google$0.17$0.75$0.00 (free tier)1M tokens
Gemini 2.0 Flash-LiteFree Tier
Google$0.07$0.30$0.00 (free tier)1M tokens
OpenRouter Free ModelsFree Tier
OpenRouterFreeFree$0.00Model-dependent
Groq (Llama 3.3 70B)Free + Fast
GroqFreeFree$0.00128K tokens
DeepSeek V3.2Budget KingHOT
DeepSeek$0.28$0.42$0.04128K tokens
GPT-4.1 MiniBudget
OpenAI$0.40$1.60$0.121M tokens
Grok 4.1 FastNewHOT
xAI$0.20$0.50$0.042M tokens
GPT-4.1Popular
OpenAI$2.00$8.00$0.601M tokens
Gemini 2.5 ProFree Tier
Google$1.25$10.00$0.681M tokens
Claude Sonnet 4.5Best Reasoning
Anthropic$3.00$15.00$1.08200K tokens
Claude Opus 4Premium
Anthropic$15.00$75.00$5.40200K tokens
GPT-4oLegacy
OpenAI$5.00$15.00$1.20128K tokens

Model Profiles

Ollama (Local)

(Your Hardware)100% Free

Requires 16GB+ RAM for 8B models. hermes-2-pro or mistral:7b recommended for tool calling.

Completely free. Runs on your CPU/GPU. No API key needed.

Gemini 2.5 Flash

(Google)Free Tier

Best free option for beginners. Massive context window. Paid tier is ultra-cheap too.

Free tier via Google AI Studio: 1,500 req/day, 1M tokens/min. No credit card.

Gemini 2.0 Flash-Lite

(Google)Free Tier

Lighter model, faster responses. Good for simple agent tasks.

Free tier via Google AI Studio. Even cheaper than 2.5 Flash on paid tier.

OpenRouter Free Models

(OpenRouter)Free Tier

Great for experimenting. Auto-router picks the best free model for your task.

30+ free models including DeepSeek, Llama 3.3, Gemma 3. 200 req/day. No credit card.

Groq (Llama 3.3 70B)

(Groq)Free + Fast

Extremely fast inference. Good for rapid agent loops. Daily request cap is the main limit.

Free tier: 1,000 req/day, 6K tokens/min. No credit card. Fastest inference (300+ tok/s).

DeepSeek V3.2

(DeepSeek)Budget King

Best price-to-performance ratio. Full tool calling. Rate limits during peak hours.

GPT-4.1 Mini

(OpenAI)Budget

Good balance of cost and capability. Reliable tool calling from OpenAI.

Grok 4.1 Fast

(xAI)New

Massive 2M context. Competitive pricing. Gaining traction fast on OpenRouter.

GPT-4.1

(OpenAI)Popular

Industry standard. Best schema adherence. Reliable for production agents.

Gemini 2.5 Pro

(Google)Free Tier

Strong reasoning with huge context. Free tier has lower limits than Flash.

Limited free tier available via Google AI Studio.

Claude Sonnet 4.5

(Anthropic)Best Reasoning

Strongest reasoning. Excellent for complex multi-step tasks. Higher cost.

Claude Opus 4

(Anthropic)Premium

Maximum capability. Only for tasks where quality justifies 50x DeepSeek cost.

GPT-4o

(OpenAI)Legacy

Previous flagship. Replaced by GPT-4.1 for most use cases.

Trending on OpenRouter

These models are seeing the most usage growth this week (Feb 2026). Many are available for free via OpenRouter.

Minimax M2.5

TRENDING

Massive adoption for programming tasks

32.6M/week

Qwen 3.5-397B

TRENDING

Multimodal MoE, strong tool calling

94.6K/week

Qwen 3-Max-Thinking

TRENDING

+199% growth, adaptive tool-use

216K/week

Kimi K2.5 (Moonshot)

TRENDING

#1 on Agent Leaderboard for tool selection

Rising

GLM-5 (Zhipu)

TRENDING

Top reasoning + coding + agent tasks

19.8M/week

Which Model Should I Use?

Zero Budget

Learning, experimenting, personal projects

  • Primary: Gemini 2.5 Flash (free tier)
  • Backup: Groq Llama 3.3 70B
  • Offline: Ollama + hermes-2-pro
Low Budget ($1-5/month)

Daily automation, personal assistant

  • Primary: DeepSeek V3.2 ($0.04/100 steps)
  • Fallback: Grok 4.1 Fast ($0.04/100 steps)
  • Heavy tasks: Gemini 2.5 Flash (paid tier)
Production ($10-50/month)

Business workflows, reliable automation

  • Primary: GPT-4.1 ($0.60/100 steps)
  • Complex reasoning: Claude Sonnet 4.5
  • Cost fallback: DeepSeek for simple tasks
Enterprise (Unlimited)

Mission-critical, safety-sensitive

  • Primary: Claude Sonnet 4.5 (best reasoning)
  • Backup: GPT-4.1 (best reliability)
  • Heavy reasoning: Claude Opus 4 (when needed)

OpenClaw Token Consumption Warning

OpenClaw agents are token-hungry. A single browser automation task can use 50-200+ steps, with each step consuming 500-2,000 tokens. Before launching long-running agents:

  • Set maxSteps in your config to prevent runaway costs
  • Use our Cost Estimator to predict spend before running
  • Start with free/budget models, upgrade only when needed
  • Be specific in prompts — vague goals = more steps = more cost

Ready to Choose Your Model?

Use the Config Wizard to generate a working OpenClaw config with your chosen model, or try the Cost Estimator to simulate your spend.

Config WizardCost Estimator

Sources & Data

OpenRouter RankingsArtificial AnalysisIntuitionLabs PricingClawKit Cost Estimator