Overview
Whim gives you access to 20+ models across three providers. You can set a default model for your workspace and override it on individual tasks. Different models offer different tradeoffs between speed, quality, context window, and CU cost.Available Models
Claude Models (Anthropic)
Available via CCR + OpenRouter and Claude Subscription.| Model | Model ID | Strengths | Best for |
|---|---|---|---|
| Claude Opus 4.6 | claude-opus-4.6 | Highest capability, deep reasoning | Complex architecture, difficult bugs, nuanced refactors |
| Claude Sonnet 4.5 | claude-sonnet-4.5 | Strong balance of speed and quality | General-purpose coding, most tasks |
| Claude Sonnet 4.5 (1M) | claude-sonnet-4.5-1m | Extended 1M token context window | Large codebases, cross-file analysis |
| Claude Haiku 4.5 | claude-haiku-4.5 | Fastest Claude model | Quick edits, simple tasks, rapid iteration |
GPT Models (OpenAI)
| Model | Model ID | Provider | Best for |
|---|---|---|---|
| GPT 5.4 | gpt-5.4 | Codex only | Flagship GPT tasks via native Codex CLI |
| GPT 5.3 | gpt-5.3 | OpenRouter | General-purpose GPT coding |
| GPT 5.3 Codex | gpt-5.3-codex | OpenRouter, Codex | Code-optimized GPT 5.3 |
| GPT 5.3 Codex Spark (Preview) | gpt-5.3-codex-spark | OpenRouter, Codex | Fast, lightweight code tasks |
| GPT 5.2 | gpt-5.2 | OpenRouter | Budget-friendly GPT option |
| GPT 5.2 Codex | gpt-5.2-codex | OpenRouter, Codex | Budget-friendly code-optimized GPT |
| GPT 5.1 Codex Mini | gpt-5.1-codex-mini | OpenRouter, Codex | Fastest/cheapest GPT option |
| GPT OSS 120B | gpt-oss-120b | OpenRouter | Open-source GPT variant |
Google Models
| Model | Model ID | Provider | Best for |
|---|---|---|---|
| Gemini 3 Pro | gemini-3-pro-preview | OpenRouter | Complex reasoning, large context |
| Gemini 2.5 Flash | gemini-2.5-flash | OpenRouter | Fast, cost-effective tasks |
Other Models
| Model | Model ID | Provider | Best for |
|---|---|---|---|
| Grok Code Fast | grok-code-fast-1 | OpenRouter | Rapid code generation |
| DeepSeek V3.2 | deepseek-v3.2 | OpenRouter | Cost-effective coding |
| MiniMax M2.5 | minimax-m2.5 | OpenRouter | General coding tasks |
| Qwen3 Coder Next | qwen3-coder-next | OpenRouter | Code-focused tasks |
| Kimi K2.5 | kimi-k2.5 | OpenRouter | General coding tasks |
Setting Your Default Model
Your default model determines which model runs when you create new tasks. You can set defaults at two levels:Workspace Default
Workspace defaults apply to all members. A workspace admin can set these in Settings > Workspace > Defaults.User Default
Your personal default overrides the workspace default. Set it in Settings > My Defaults or during onboarding. Each provider has its own default model:- CCR + OpenRouter: Claude Sonnet 4.5
- Claude Subscription: Claude Opus 4.6
- Codex Subscription: GPT 5.4
Per-Task Model Override
You can override the default model when creating any task. In the task creation dialog, select a different model from the model dropdown. This override only affects that specific task — your default stays the same.CU Cost Considerations
How models affect your CU usage depends on your provider:CCR + OpenRouter
CU cost has two components:- Container runtime — charged at 1 CU per 30 minutes regardless of model
- API tokens — varies by model; more capable models cost more tokens per CU
Claude Subscription / Codex Subscription
With subscription providers, Whim only charges CU for container runtime (1 CU per 30 minutes). Token costs go through your existing Anthropic or OpenAI subscription. This makes subscription providers more CU-efficient for token-heavy tasks.Model Selection Tips
Start with the default, then adjust
Start with the default, then adjust
The default model for each provider is a good general-purpose choice. Try it first and switch if you need more speed or capability.
Use flagship models for complex tasks
Use flagship models for complex tasks
Claude Opus 4.6 and GPT 5.4 are best for tasks that require deep reasoning — complex debugging, large refactors, or architectural decisions.
Use lighter models for simple tasks
Use lighter models for simple tasks
Claude Haiku 4.5, Gemini 2.5 Flash, and GPT 5.1 Codex Mini are fast and cost-effective for simple edits, formatting, boilerplate, and quick fixes.
Use extended context for large codebases
Use extended context for large codebases
Claude Sonnet 4.5 (1M context) is ideal when the agent needs to reason across many files simultaneously.
Next Steps
Fast Mode & Reasoning
Fine-tune speed vs quality with fast mode and reasoning effort.
Compute Units
Understand how CU usage is calculated.