Appearance
Supported Models
Pencel connects to three AI providers: Anthropic, Google, and OpenAI. Each model has different strengths, speeds, and costs. You choose which model an agent uses, and you can mix models across agents in the same workspace.
Available Models
| Provider | Model | Model ID | Best For | Speed | Relative Cost |
|---|---|---|---|---|---|
| Anthropic | Claude Opus 4.6 | claude-opus-4-6 | Complex reasoning, nuanced writing, multi-step analysis | Moderate | $$$ |
| Anthropic | Claude Sonnet 4.6 | claude-sonnet-4-6 | Balanced quality and speed, default model | Fast | $$ |
| Anthropic | Claude Haiku 4.5 | claude-haiku-4-5-20251001 | Quick tasks, classification, high-volume processing | Very Fast | $ |
| Anthropic | Claude 3.5 Sonnet | claude-3-5-sonnet-20241022 | Strong reasoning, legacy compatibility | Moderate | $$ |
| Anthropic | Claude 3.5 Haiku | claude-3-5-haiku-20241022 | Fast lightweight tasks, legacy compatibility | Fast | $ |
| Anthropic | Claude 3 Opus | claude-3-opus-20240229 | Deep analysis, legacy compatibility | Slow | $$$ |
| Anthropic | Claude 3 Haiku | claude-3-haiku-20240307 | Ultra-fast simple tasks, legacy compatibility | Very Fast | $ |
| Gemini 2.5 Flash | gemini-2.5-flash | Fast routine tasks, research, summarization | Fast | $ | |
| Gemini 2.5 Pro | gemini-2.5-pro | Complex analysis, long context windows, detailed reports | Moderate | $$ | |
| OpenAI | GPT-4o | — | General purpose, strong reasoning, broad knowledge | Moderate | $$ |
| OpenAI | GPT-4o mini | — | Simple tasks, classification, lightweight processing | Fast | $ |
INFO
The default model for new agents is Claude Sonnet 4.6 (claude-sonnet-4-6). You can change this per agent in the agent editor.
Managed vs. BYOK availability
On a Pro Desktop subscription, these models are available through Pencel-managed credits with no API key required: all current Claude 4.x models (claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5-20251001) and Claude 3.5 (claude-3-5-sonnet, claude-3-5-haiku), gpt-4o and gpt-4o-mini, and Gemini 2.5 Flash and Gemini 2.5 Pro.
Models not in that set (e.g. the legacy Claude 3 Opus/Haiku) are BYOK-only — add your own provider key in Settings → LLM Providers to use them. On the Free plan, every model is BYOK.
The model picker also offers two managed modes on a subscription: Pencel Auto (cost-optimized — Pencel picks the model) and Pencel Private (zero data retention). See Accounts & sign-in.
Choosing the Right Model
Pick a model based on what the agent needs to do, not on the provider name.
Use Claude Opus 4.6 when:
- The task requires careful, nuanced writing (executive briefings, customer communications)
- The agent needs to follow complex, multi-layered instructions
- Accuracy is more important than speed
- The output is customer-facing or goes to leadership
Use Claude Sonnet 4.6 (default) when:
- You want the best balance of quality, speed, and cost
- The task is moderately complex (reports, analysis, multi-step workflows)
- You are unsure which model to pick
Use Gemini 2.5 Flash when:
- The task is primarily research or data gathering
- You need to process many items quickly
- Cost efficiency matters (high-volume workflows)
- The task is routine and well-defined
Use Gemini 2.5 Pro when:
- You need to analyze very long documents (up to ~1M tokens of context)
- The task requires deep analysis but Claude's cost is prohibitive
- You want a balance between quality and price
Use GPT-4o when:
- You want strong general-purpose reasoning
- The task involves broad world knowledge
- You need a good balance of quality and speed
Use Claude Haiku 4.5, GPT-4o mini, or Gemini 2.5 Flash when:
- The task is simple (classification, extraction, formatting)
- You want the lowest possible cost
- Speed matters more than depth
TIP
A common pattern is to use a cheaper model for data gathering steps and a premium model for the final synthesis step. You can assign different agents to different workflow steps, each with their own preferred model.
Setting a Model for an Agent
- Open the agent you want to configure (or create a new one).
- Find the Preferred Model field.
- Select a model from the dropdown.
- Save the agent.
Every time this agent runs — whether in chat or as part of a workflow — it uses the selected model.
How Model Routing Works
Pencel automatically routes requests to the right provider based on the model you selected. You don't need to think about API endpoints or provider-specific settings. Just pick the model and go.
INFO
Each provider requires its own API key. Set up your keys in Settings before using models from that provider. You only need keys for the providers you actually use.
Circuit Breakers and Provider Health
Pencel monitors the health of each provider in real time. If a provider starts returning errors (network issues, rate limits, outages), the circuit breaker activates:
- Closed (healthy) — Requests flow normally.
- Open (unavailable) — The provider is temporarily disabled. Pencel stops sending requests to avoid cascading failures.
- Half-open (testing) — Pencel periodically tests whether the provider has recovered.
You can see the current status of all providers in the diagnostics view on the Jobs page (/jobs).
WARNING
When a provider's circuit breaker is open, agents using that provider's models cannot run. They will wait for the provider to recover. Pencel does not automatically fall back to a different provider — the agent uses the model you assigned.
Cost Tracking
Every model call is tracked in the Cost & Usage section of Settings. You can see:
- Per-run costs — How much each workflow run consumed
- Per-date costs — Daily spending across all agents
- Input vs. output tokens — Input tokens (what you send) are cheaper than output tokens (what the model generates)
The relative cost column in the table above gives a rough guide:
| Symbol | Meaning |
|---|---|
| $ | Low cost — suitable for high-volume use |
| $$ | Moderate cost — good for important tasks |
| $$$ | Premium cost — reserve for high-value outputs |
TIP
If your monthly costs are climbing, the Cost & Usage section is the best place to start. Look for agents running expensive models on routine tasks, and consider switching them to a cheaper alternative.
API Key Requirements
| Provider | Where to Get a Key |
|---|---|
| Anthropic | console.anthropic.com |
| aistudio.google.com | |
| OpenAI | platform.openai.com |
Keys are stored securely in your operating system's keychain. They never leave your machine except to authenticate with the respective provider. See Settings for setup instructions.