Model availability and IDs checked on October 14, 2026 against the haimaker catalog. These are candidates to test on your own tasks, not results from a new controlled benchmark.
Everyone asks which model to use. The honest answer: it depends on what you’re doing and how much you want to spend.
Note: Clawdbot has been rebranded to OpenClaw — same powerful AI agent platform, new name. Learn more at openclaw.ai.
OpenClaw supports a dozen providers. Anthropic, OpenAI, Google, open-source models through haimaker.ai. Each has tradeoffs around cost, capability, and where your data ends up.
Here’s how I think about picking one.
OpenClaw supported models list
OpenClaw can use any model exposed through a supported provider or an OpenAI-compatible endpoint. The practical supported list is wider than the default config.
| Provider | Current candidates | Workload to test |
|---|---|---|
| Haimaker | MiniMax M3, Kimi K3, GLM-5.3 Flash, DeepSeek V4.1 Flash | Compare cloud models with one API key |
| Anthropic | Sonnet 5.5, Opus 5.5, Haiku 5.5 | Daily work, harder coding, and focused tasks |
| OpenAI | GPT-6.1 Sol, GPT-6 Luna | Coding and bounded high-volume work |
| Gemini 3.8 Flash, Gemini 4 Argon, Flash-Lite | Daily tasks, difficult research, lighter processing | |
| xAI | Grok 4.6, Grok 4.1 Fast | Compare reasoning with large-context budget work |
| DeepSeek | V4 Pro, V4.1 Flash | Coding and budget cloud alternatives |
| Ollama | Qwen3.8, Muse Glimmer, Qwen3 Coder, Gemma 4 | Local inference on suitable hardware |
If the model is OpenAI-compatible, the rule is simple: add the provider, add the model ID, then allowlist the fully qualified model name. That covers Haimaker, Ollama, OpenRouter, Together, private vLLM deployments, and most hosted inference gateways.
Price, capability, privacy
These three things compete with each other. You can optimize for two, maybe, but rarely all three.
Price
Token rates vary by model, endpoint, caching, and service tier. Compare both input and output prices in the live catalog. Claude Sonnet 5.5 and GPT-6.1 Sol are current mid-range candidates; Opus 5.5 is a more expensive option for difficult work. Budget models deserve a comparison when the task is routine.
The useful measure is completed-task cost. Count failed tool calls and retries alongside tokens, then compare the result with a stronger model.
Capability
Benchmarks lie. For OpenClaw, what actually matters:
- Tool calling – Can it invoke shell commands and APIs without fumbling the syntax?
- Context tracking – Does it remember what you said 50 messages ago?
- Code quality – When it writes code, does it run?
- Speed – How long before it starts responding?
Privacy
Cloud APIs mean your prompts hit external servers. For personal finance, health data, or proprietary code, that’s a problem. You can self-host open-source models, but that requires hardware and tolerance for latency.
Recommendations by use case
Daily assistant work
Claude Sonnet 5.5 is a current starting point for messages, files, and multi-step automation. Compare it with Claude Haiku 5.5, GPT-6 Luna, or GLM-5.3 Flash on routine tasks before choosing a paid default.
Coding and automation
Test Claude Opus 5.5 or GPT-6.1 Sol on difficult debugging and multi-file changes. Anthropic’s model lineup and OpenAI’s Sol guide describe their intended roles. Validate the generated patch and run your tests; a newer model does not remove that work.
Research and document analysis
Gemini 4 Argon is a new paid candidate for difficult analysis. Gemini 3.8 Flash is another large-context option. Verify answers against the source documents and check the endpoint’s output limit before asking for long reports.
Gemini models in OpenClaw
For a new setup, compare Gemini 3.8 Flash for everyday work with Gemini 4 Argon for harder reasoning. Gemini 3.5 Flash-Lite is worth trying on lighter processing. Google’s current model list replaces the old Pro-first recommendation.
Google’s free tier is available for selected Flash models through Google directly, subject to your project quota. Argon is paid-only, and gateway calls have separate billing.
To add Gemini models through haimaker.ai, add them to your provider config:
{
models: {
providers: {
haimaker: {
// ... existing config
models: [
// ... existing models
{ id: "google/gemini-4-argon", name: "Gemini 4 Argon" },
{ id: "google/gemini-3.8-flash", name: "Gemini 3.8 Flash" }
]
}
}
}
}
Or use them directly through Google’s API if you prefer. OpenClaw supports Google as a first-party provider.
Running models locally with OpenClaw
Qwen3.8 27B and Muse Glimmer 30B are newer local candidates with tools support. Their Ollama downloads are about 18GB before context and runtime overhead, so leave memory headroom. For smaller machines, test Gemma 4 E4B or Qwen3 8B.
ollama pull qwen3.8:27b
{
models: {
providers: {
ollama: {
baseUrl: "http://localhost:11434/v1",
api: "openai-completions",
models: [{ id: "qwen3.8:27b", name: "Qwen3.8 27B", contextWindow: 32768, maxTokens: 8192 }]
}
}
},
agents: { defaults: { model: { primary: "ollama/qwen3.8:27b" } } }
}
The 32K context here is a starting configuration, not a guarantee that every machine fits it. See the local model guide for memory tiers or Gemma 4 setup for a smaller-model walkthrough.
Privacy-sensitive work
Use Ollama or another local runtime when the prompts must stay on your machine. Open weights hosted through a cloud API still send prompts to external infrastructure. If you use a gateway, check the actual provider’s data handling and routing terms before sending sensitive information.
Keep cloud and local configurations separate, and choose the endpoint as well as the model for each task.
Provider comparison
Anthropic (Claude)
Sonnet 5.5, Opus 5.5, and Haiku 5.5 cover different cost and workload tiers. Start with a representative tool sequence and compare the total result rather than assuming every task needs Opus.
OpenAI (GPT)
GPT-6.1 Sol is a current coding and agent candidate. GPT-6 Luna targets focused, high-volume work. Use exact model IDs and verify the limits of the endpoint you choose.
Google (Gemini)
Gemini 3.8 Flash and Gemini 4 Argon are current large-context candidates. Flash-Lite covers lighter work. Direct Google free-tier access and gateway billing are different services.
Open models through haimaker.ai
MiniMax M3, Kimi K3, GLM-5.3 Flash, and DeepSeek V4.1 Flash are available in the catalog. Hosted open models still use cloud infrastructure; check provider terms when privacy is part of the decision.
The API is OpenAI-compatible:
curl https://api.haimaker.ai/v1/chat/completions \
-H "Authorization: Bearer $HAIMAKER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax/minimax-m3",
"messages": [{"role": "user", "content": "Hello!"}]
}'
Swap your base URL to https://api.haimaker.ai/v1 and you’re done.
GET $10 FREE CREDITS ON HAIMAKER
OpenClaw configuration
Setting your default model in ~/.openclaw/openclaw.json:
{
agents: { defaults: { model: { primary: "anthropic/claude-sonnet-5-5" } } }
}
Switch models mid-session with /model opus or /model haimaker/minimax/minimax-m3.
Migration note: If you’re coming from Clawdbot, your config files are automatically migrated. The CLI now uses
openclawinstead ofclawdbotcommands.
Adding haimaker.ai as a provider in OpenClaw
{
env: { HAIMAKER_API_KEY: "sk-..." },
agents: {
defaults: { model: { primary: "haimaker/minimax/minimax-m3" } }
},
models: {
mode: "merge",
providers: {
haimaker: {
baseUrl: "https://api.haimaker.ai/v1",
apiKey: "${HAIMAKER_API_KEY}",
api: "openai-completions",
models: [
{ id: "minimax/minimax-m3", name: "MiniMax M3" },
{ id: "deepseek/deepseek-v4.1-flash", name: "DeepSeek V4.1 Flash" },
{ id: "z-ai/glm-5.3-flash", name: "GLM-5.3 Flash" }
]
}
}
}
}
Prefer to skip the hand-editing? npx -y @haimaker/connect --openclaw writes that exact provider block into ~/.openclaw/openclaw.json for you; the connect guide covers the other agents it handles too.
Which model should I choose?
Choose by the job you need to finish:
- Daily automation: compare Claude Sonnet 5.5 with a budget candidate
- Hard coding: test Claude Opus 5.5 or GPT-6.1 Sol
- Difficult research: compare Gemini 4 Argon with Gemini 3.8 Flash
- No API token charge: run a model locally through Ollama
- Budget cloud work: compare MiniMax M3, DeepSeek V4.1 Flash, and GLM-5.3 Flash
- Private data: choose a local endpoint when the data must remain on your machine
The free models guide separates local inference from cloud free tiers. Keep a stronger fallback for tasks the cheaper model cannot finish, and measure your own split.
OpenClaw models list
Current shortlist by provider. This is not an exhaustive supported-model inventory; use the live catalog for exact IDs and availability.
Anthropic: Claude Sonnet 5.5 · Opus 5.5 · Haiku 5.5 · Fable 5.1
OpenAI: GPT-6.1 Sol · GPT-6 Astra · GPT-6 Luna
Google: Gemini 4 Argon · Gemini 3.8 Flash · Gemini 3.5 Flash-Lite
xAI: Grok 4.6 · Grok 4.1 Fast
Budget/open-model candidates: DeepSeek V4.1 Flash · DeepSeek V4 Pro · MiniMax M3 · GLM-5.3 Flash · Kimi K3 · Qwen3.7 Plus
Local Ollama candidates: Qwen3.8 27B · Muse Glimmer 30B · Qwen3 Coder 30B · gpt-oss 20B · Gemma 4 E4B
Bottom line
Start with one model suited to your most frequent task and compare it with one alternative. Record correct completion, failed tool calls, retries, elapsed time, and total cost. Add another model only when it solves a failure you actually encounter.
Ready to set up your own OpenClaw agent? Visit openclaw.ai to get started or check out the OpenClaw documentation for detailed configuration options.