OpenClaw API Keys: Which LLM Provider to Choose? (2026)

OpenClaw supports multiple LLM providers, and choosing the right one significantly affects your agent's quality, speed, and cost. This guide compares the three major providers — Anthropic, OpenAI, and Google — with current pricing, quality analysis, and recommendations for different use cases.

The Three Major Providers

Anthropic (Claude)

Anthropic makes Claude, widely regarded as the best model for agentic tasks in 2026. Claude excels at following complex instructions, writing code, and maintaining context across long conversations.

Available Models:

Why choose Anthropic:

• Claude Opus 4 is the best model for complex agent tasks
• Claude Sonnet 4 offers the best quality-to-price ratio
• Excellent instruction following — critical for SOUL.md adherence
• 200K context window is generous
• Constitutional AI approach reduces harmful outputs

Setup:

# Get your API key from https://console.anthropic.com
openclaw config set llm.provider anthropic
openclaw config set llm.api_key "sk-ant-your-key-here"
openclaw config set llm.model "claude-sonnet-4-20250514"

OpenAI (GPT)

OpenAI is the most well-known provider and offers competitive pricing, especially at the lower tiers.

Available Models:

Why choose OpenAI:

• GPT-4o mini is extremely cheap for simple tasks
• Largest ecosystem of tools and integrations
• Function calling is well-implemented
• Most documentation and community resources

Setup:

openclaw config set llm.provider openai
openclaw config set llm.api_key "sk-your-key-here"
openclaw config set llm.model "gpt-4o"

Google (Gemini)

Google offers the most generous free tier and competitive pricing at scale.

Available Models:

Why choose Google:

• Free tier for experimentation and light use
• 1M token context window — no other provider comes close
• Gemini Flash is the cheapest option per token
• Strong multimodal capabilities (image, video, audio)

Setup:

openclaw config set llm.provider google
openclaw config set llm.api_key "AIza-your-key-here"
openclaw config set llm.model "gemini-2.5-pro"

Quality Comparison

Price is only half the equation. Here is how the models compare in quality for typical OpenClaw tasks:

Instruction Following

How well does the model adhere to your SOUL.md rules?

Code Generation

Tool Use / Skill Execution

Long Context Retention

Cost Scenarios

Scenario 1: Light Personal Use (500K tokens/month)

Recommendation: Google Gemini free tier or Flash for budget users. Claude Haiku if you want better quality.

Scenario 2: Daily Power Use (5M tokens/month)

Recommendation: Claude Sonnet for quality, Gemini Flash for budget.

Scenario 3: Heavy Business Use (20M tokens/month)

Recommendation: Use model routing — Claude Sonnet for complex tasks, Gemini Flash for simple ones.

Setting Up Multiple Providers

One of OpenClaw's strengths is multi-provider support. Configure multiple providers and route requests based on task complexity:

llm:
  providers:
    anthropic:
      api_key: "${ANTHROPIC_API_KEY}"
      models:
        - claude-sonnet-4-20250514
        - claude-opus-4-20250514
    openai:
      api_key: "${OPENAI_API_KEY}"
      models:
        - gpt-4o
        - gpt-4o-mini
    google:
      api_key: "${GOOGLE_API_KEY}"
      models:
        - gemini-2.5-pro
        - gemini-2.5-flash

  routing:
    default: "claude-sonnet-4-20250514"
    simple_queries: "gemini-2.5-flash"
    complex_reasoning: "claude-opus-4-20250514"
    code_generation: "claude-sonnet-4-20250514"
    
  fallback:
    - "gpt-4o"  # If primary fails, try this
    - "gemini-2.5-pro"  # Then this

This setup gives you the best quality where it matters and saves money on routine tasks.

How to Get API Keys

Anthropic

1. Go to console.anthropic.com
2. Create an account
3. Add a payment method (required even for free credits)
4. Go to API Keys and create a new key
5. Copy the key — it starts with sk-ant-

New accounts get $5 in free credits.

OpenAI

1. Go to platform.openai.com
2. Sign up or log in
3. Go to API Keys in the sidebar
4. Create a new secret key
5. Copy the key — it starts with sk-

New accounts get $5 in free credits for the first 3 months.

Google

1. Go to aistudio.google.com
2. Sign in with your Google account
3. Click "Get API Key"
4. Create a key in a new or existing project
5. Copy the key — it starts with AIza

Free tier available with rate limits. No credit card required.

Switching Providers

If you want to switch your primary provider:

# Change provider and model
openclaw config set llm.provider anthropic
openclaw config set llm.model "claude-sonnet-4-20250514"
openclaw config set llm.api_key "sk-ant-new-key"

# Restart to apply
openclaw restart

# Verify
openclaw config get llm

Your conversation history and memory persist across provider switches. The agent may behave slightly differently with a new model, so update your SOUL.md if needed.

API Key Security

Store Keys Securely

Never put API keys directly in config.yaml. Use environment variables:

# In .env file (chmod 600)
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GOOGLE_API_KEY=AIza...

# In config.yaml, reference variables
llm:
  api_key: "${ANTHROPIC_API_KEY}"

Set Spending Limits

Every provider allows spending limits. Set them immediately after creating your key:

• Anthropic: Console > Settings > Limits
• OpenAI: Platform > Settings > Limits
• Google: Cloud Console > Billing > Budgets

Rotate Keys Periodically

Change your API keys every 90 days:

# Generate new key in provider dashboard
# Update your .env
nano ~/.openclaw/.env

# Restart
openclaw restart

Monitor Usage

Set up spending alerts via email. If your agent's token usage spikes unexpectedly, you want to know immediately. Use Inbounter to send yourself spending alerts:

openclaw cron add --name "cost-alert" "0 * * * *" \
  "Check today's API spending. If it exceeds $5, send an alert 
   email via Inbounter to admin@company.com with the current 
   spending amount and breakdown by model."

Our Recommendation

For most OpenClaw users in 2026:

1. Start with Claude Sonnet 4 as your primary model. It offers the best balance of quality, speed, and cost for agentic tasks.

2. Add Gemini Flash as your budget model for simple queries and cron jobs.

3. Keep Claude Opus 4 available for complex reasoning tasks that justify the premium price.

4. Use model routing to automatically direct queries to the right model based on complexity.

This multi-provider setup gives you top-tier quality where it matters while keeping monthly costs reasonable.

Frequently Asked Questions

Can I use a local LLM with OpenClaw?

Yes. OpenClaw supports Ollama and llama.cpp. Set llm.provider: ollama and point to your local instance. Quality will be lower than cloud APIs but cost is zero after hardware investment.

Which provider has the best uptime?

All three major providers have 99.9%+ uptime. Configure fallback providers so if one goes down, your agent automatically switches to another.

Do I need separate API keys for each model?

No. Each API key gives you access to all models available on your tier within that provider.

Can I use Claude for free?

Claude does not have a free API tier. Anthropic gives $5 in initial credits for new accounts. After that, you pay per token. Google Gemini is the only major provider with an ongoing free tier.

How do I check my current API spending?

openclaw costs estimate  # Shows estimated spending
# Or check directly in each provider's dashboard

Will switching providers affect my agent's personality?

Slightly. Different models interpret SOUL.md differently. After switching, test your agent and adjust SOUL.md wording if the behavior changes in unwanted ways.

Need your OpenClaw agent to send emails or SMS? Inbounter provides communication APIs designed for AI agents. Works with any LLM provider.