π§ Gemini Model Guide
Orchable supports a wide range of Gemini models, each optimized for different stages of the orchestration pipeline.
π Model Comparisonβ
| Model | Primary Use Case | Context Window | Key Features |
|---|---|---|---|
| Gemini 3 Pro | Highest reasoning, complex coding | 1M+ tokens | State-of-the-art reasoning, SOTA vibe-coding |
| Gemini 3 Flash | Balanced speed & intelligence | 1M+ tokens | Pro-grade reasoning at Flash speed |
| Gemini 2.5 Pro | Deep STEM, math, long context | 1M+ tokens | Optimized for analytical depth |
| Gemini 2.5 Flash | Production workhorse | 1M+ tokens | Best price-performance; ultra-reliable |
| Gemini 2.5 Flash Lite | High-volume micro-tasks | 1M+ tokens | Fastest, most affordable |
| Gemini 2.0 Flash | Legacy support / Stable Gen | 1M tokens | Mature, stable performance |
π οΈ Capability Matrixβ
| Capability | Gemini 3 Series | Gemini 2.5 Series | Gemini 2.0 Series |
|---|---|---|---|
| Structured Output (JSON) | β | β | β |
| Thinking / CoT | β (Multi-level) | β (Budget) | β (Exp) |
| Context Caching | β | β | β |
| Code Execution | β | β | β |
| Multimodal (Audio/Video) | β | β | β |
| Live API (Streaming) | β | β | β |
π Choosing a Modelβ
- Planning Stage: Use Gemini 3 Flash or Gemini 3 Pro for high-quality orchestrations.
- Processing Stage: Use Gemini 2.5 Flash for reliable, consistent output.
- Filtering Stage: Use Gemini 2.5 Flash Lite to save cost and time.
- RAG Stage: Use Gemini Embedding 1 for all vector retrieval tasks.
Last Updated: 2026-02-24