OpenWorker Supported Models

A practical comparison of every AI model you can run with OpenWorker — performance, cost, and best use cases for each provider.

See Model Comparison ← Back to Guide

Cloud Models for OpenWorker

These models run on provider infrastructure and deliver the fastest, most capable results inside OpenWorker. You pay per token through your own API key.

Model Provider Context Tool Calling Speed Cost Best For
GPT-5.6 Sol OpenAI 128K Excellent Fast $15 / 1M tokens Complex reasoning, multi-step planning, detailed analysis
GPT-5.6 Sol Mini OpenAI 128K Very Good Very Fast $4 / 1M tokens Quick tasks, drafting, summarization
Claude Fable Anthropic 200K Excellent Fast $12 / 1M tokens Nuanced writing, careful reasoning, safety-sensitive tasks
Claude Sonnet 4.6 Anthropic 200K Very Good Very Fast $3 / 1M tokens Balanced speed and quality, coding tasks
Gemini 3.6 Google 1M Very Good Fast $10 / 1M tokens Multimodal tasks, very long documents, image analysis
Gemini 3.6 Flash Google 1M Good Very Fast $1.5 / 1M tokens High-volume tasks, cost-sensitive workflows
DeepSeek R2 DeepSeek 128K Good Medium $2 / 1M tokens Coding, technical analysis, budget-friendly reasoning
Kimi K3 Moonshot 256K Good Medium $3 / 1M tokens Long document processing, research synthesis

Local Models via Ollama for OpenWorker

Run these models locally through OpenWorker and Ollama on your own Mac. Zero API cost, complete privacy, and no internet connection required. Trade-off is slower speed and lower quality compared to premium cloud models.

Model RAM Required Quality Speed Best For
Llama 3.3 70B 40 GB Near GPT-4 level Slow on consumer hardware Privacy-critical tasks, offline use
Llama 3.3 8B 5 GB Good for simple tasks Fast on most Macs Quick drafts, email triage, basic summarization
Qwen 3 32B 20 GB Strong reasoning Medium Multilingual tasks, coding, analysis
Mistral Large 2 24 GB Very capable Medium Business writing, European language support
DeepSeek R2 Lite 8 GB Solid reasoning Medium Code generation, technical documentation

Which Model Should You Use with OpenWorker?

The best model depends on your task, budget, and privacy requirements. Here are our recommendations for common OpenWorker workflows.

Daily office work (email, calendar, Slack)

Recommended: GPT-5.6 Sol Mini or Gemini 3.6 Flash

Fast response times keep your OpenWorker workflow smooth. These models handle routine OpenWorker tasks well at low cost. A typical day of email triage and calendar management costs under $0.50.

Customer-facing writing (briefs, reports, proposals)

Recommended: Claude Fable or GPT-5.6 Sol

Premium models produce more polished, nuanced writing in OpenWorker. The extra cost per task is negligible when the output goes directly to clients or stakeholders.

Code and technical documentation

Recommended: DeepSeek R2 or Claude Sonnet 4.6

Both excel at understanding code in OpenWorker. DeepSeek is more cost-effective for OpenWorker coding workflows; Claude Sonnet offers faster iteration cycles.

Sensitive or confidential data

Recommended: Ollama with Llama 3.3 8B or Qwen 3 32B

Local models in OpenWorker ensure nothing leaves your machine. Even your prompts stay private. The quality trade-off is worth it when handling legal documents, HR data, or financial records.

Processing very long documents (100+ pages)

Recommended: Gemini 3.6 (1M context) or Kimi K3 (256K)

These models have the largest context windows, allowing OpenWorker to process entire documents without chunking. Essential for contract review, research papers, and audit reports.

OpenWorker Model Selection Strategy

The single most important factor in model selection for OpenWorker is tool-calling reliability. Unlike simple chat, OpenWorker needs the model to correctly decide when to call integrations, what parameters to pass, and how to chain multiple tool calls together. Models with excellent tool-calling scores — GPT-5.6 Sol, Claude Fable, and Claude Sonnet 4.6 — produce the most reliable multi-step workflows.

A practical approach for OpenWorker is to use a fast, cheap model as your default and switch to a premium model for high-stakes tasks. Set GPT-5.6 Sol Mini or Gemini 3.6 Flash as your daily OpenWorker driver for email triage, calendar checks, and quick summaries. When you need OpenWorker to prepare a client-facing brief or analyze a complex document, switch to GPT-5.6 Sol or Claude Fable for that specific task.

For teams handling sensitive data, the hybrid approach in OpenWorker works well: use local Ollama models for internal document processing and classification, then switch to a cloud model only when the OpenWorker task requires higher quality output that will be reviewed by a human before sharing externally. This minimizes data exposure while maintaining output quality where it matters most.

OpenWorker Models FAQ

Can I switch models in the middle of an OpenWorker task?
Yes. OpenWorker lets you change the active model between tasks without restarting. Within a single multi-step task, the agent uses the model you selected at the start, but you can change it for the next request immediately.
Which model is cheapest for OpenWorker?
For cloud models, Gemini 3.6 Flash is the most cost-effective at roughly $1.50 per million tokens. For zero cost, install Ollama and run a local model like Llama 3.3 8B. The trade-off is slower response times and lower quality compared to premium cloud models.
Does OpenWorker support fine-tuned models?
OpenWorker supports any model accessible through its supported providers. If you have a fine-tuned model hosted on OpenAI or another compatible API, you can point OpenWorker to it by specifying the model ID in settings.
Why do some models work better with OpenWorker than others?
OpenWorker relies heavily on tool calling — the ability for a model to decide when and how to use integrations. Models with strong tool-calling capabilities like GPT-5.6 Sol and Claude Fable produce more reliable multi-step workflows than models with weaker tool support.
Can I use multiple API keys from the same provider in OpenWorker?
Yes. You can add multiple keys per provider in OpenWorker settings. This is useful for teams that want to separate billing across projects or departments while sharing the same OpenWorker installation.
How much does a typical day of OpenWorker usage cost?
For a knowledge worker processing 20-30 tasks daily with a mid-tier model like Claude Sonnet 4.6 or GPT-5.6 Sol Mini, expect to spend between $0.50 and $2.00 per day. Heavy users running complex multi-step workflows with premium models may spend $5-10 per day.
What happens if my API key runs out of credits?
OpenWorker will show an error message from the provider and pause the current task. It will not switch to another model automatically. You can add credits to your provider account and retry, or switch to a different configured model to continue working.
Are Ollama models slower than cloud models in OpenWorker?
Generally yes. Cloud models run on dedicated GPU clusters optimized for inference speed. Local Ollama models depend on your Mac's hardware. An M3 Pro can run a 7B model comfortably, but larger models like 70B will be noticeably slower than their cloud equivalents.

Ready to Pick Your Model?

Install OpenWorker, add your preferred API key, and start with a mid-tier model. You can always switch later as you learn which workflows benefit from premium models.

Setup Guide Compare Alternatives