Model selection
Filo Agent lets you switch between different models within the same workflow: use faster, more efficient models for lightweight tasks, and switch to stronger reasoning, longer context, or multimodal models for more complex work.
Quick selection
Start with the kind of work you are doing, then pick the model that matches the need.
| Task type | Model | Best for |
|---|---|---|
| Everyday email and Q&A | Claude Haiku 4.5 | Quick summaries, everyday drafts, simple questions |
| General complex work | Claude Sonnet 4.6 | Complex emails, long documents, cross-tool analysis |
| Advanced reasoning | GPT-5.5 / GPT-5.4 | Planning, critical review, professional writing, coding help |
| Long documents, multimodal work, and large projects | Gemini / DeepSeek / GLM | Long context, images, videos, screenshots, large codebases |
Model comparison
Token usage is a relative level. Final usage can vary based on context length, cache hits, and output length.
| Model | Provider | Token usage | Best for |
|---|---|---|---|
Claude Haiku 4.5 | Anthropic | Lower | Fast everyday summaries, short reply drafts, lightweight Q&A, and low-stakes inbox cleanup. |
Claude Sonnet 4.6 | Anthropic | Medium | Complex email workflows, document analysis, cross-tool research, and dependable default agent work. |
GPT-5.4 | OpenAI | Medium | Professional writing, structured reasoning, code help, and tasks that need careful judgment. |
GPT-5.5 | OpenAI | Highest | High-stakes planning, deep code work, product review, and complex multi-step decisions. |
DeepSeek V4 Pro | DeepSeek | Lowest | Large codebases, automation-heavy tasks, technical synthesis, and cost-sensitive long runs. |
GLM 5.2 | Z.ai | Lowest | Engineering workflows, tool-heavy execution, long-running agent tasks, and structured operations. |
Gemini 3.1 Pro Preview | Medium | PDF understanding, image or screenshot analysis, research workflows, and multimodal review. | |
Gemini 3.5 Flash | Lowest | Fast long-context reading, media-heavy inputs, broad exploration, and parallel agent work. |
How to choose
Use the fastest capable model for routine work, then move up when the task needs deeper reasoning, more context, or multimodal understanding.