AI Models
The three AI tiers of Pollo Assistance and the multi-AI brain system.
How it works: the Brain Orchestrator
Pollo Assistance uses a multi-AI architecture. Instead of being locked to a single model, each tier (Palos, Luces, Summum) has a brain orchestrator — an AI that analyzes every message you send and decides which AI from its pool is best suited to answer it.
The brain also decides automatically when to:
- Search the web — when it needs up-to-date information, it performs a live search and shows a Sources button with the pages consulted.
- Draw an interactive diagram — complex explanations are rendered as interactive mind maps inside the chat.
- Activate think mode — for difficult tasks, the brain shows a collapsible panel with its internal reasoning before the response (Luces and Summum only).
- Open a chess game — when you ask to play chess, a playable board appears inline in the chat.
- Generate images — the brain can create images from text descriptions using Pollinations AI.
Important: Once you send the first message in a chat, the tier is locked for that conversation. If you want to switch tiers, create a new chat.
Available AIs in the Pool
The brain orchestrator picks from the following AI models depending on what each message needs:
| AI Model | Provider | Parameters | Speed | Best for |
|---|---|---|---|---|
| Llama 3.1 8B | Meta | 8B | Very fast | Quick chat, translations, simple questions |
| Llama 3.3 70B | Meta | 70B | Fast | Code, analysis, math, reasoning, creative writing |
| Llama 4 Scout | Meta | 17B (16 experts) | Fast | Vision (image understanding), multilingual, code, reasoning |
| Qwen 3 32B | Alibaba | 32B | Medium | Code, math, science, multilingual, reasoning |
| GPT OSS 20B | OpenAI | 20B | Medium | General chat, creative writing, analysis |
| GPT OSS 120B | OpenAI | 120B | Slower | Deep analysis, complex reasoning, science, code |
| Kimi K2 | Moonshot | Large | Slower | Deep analysis, complex reasoning, math, science |
All models run through the Groq infrastructure, which provides very low response times thanks to its specialized hardware (LPU — Language Processing Unit).
Image AI
In addition to text models, all tiers have access to:
- Pollinations AI — generates images from text descriptions (text-to-image).
- Llama 4 Scout (Vision) — analyzes images you upload and answers questions about them.
Pollo Palos v1.0
| Brain model | Llama 3.1 8B |
|---|---|
| AI pool | Llama 3.1 8B |
| Availability | Free |
| Max tokens | ~600 per response |
| Streaming | No |
| Features | Chat, voice |
| Best for | Quick responses and simple questions |
Pollo Palos is the lightest and fastest tier. Its brain and pool consist of a single fast model (Llama 3.1 8B with 8 billion parameters), delivering near-instant responses. Ideal for short questions, quick translations, simple calculations, or anything that doesn't require deep analysis.
Because it uses a smaller model, responses are more concise and it may struggle with very complex tasks. It's the perfect choice when you need something fast and to the point.
Pollo Luces v1.0
| Brain model | Llama 3.3 70B |
|---|---|
| AI pool | Llama 8B, Llama 70B, Llama 4 Scout, Qwen 3 32B, GPT OSS 20B |
| Availability | Free |
| Max tokens | Up to 2,000 per response (depends on AI selected) |
| Streaming | Yes (progressive response) |
| Features | Chat, voice, images, code, chess |
| Best for | Daily use, normal conversations |
Pollo Luces is the recommended tier for daily use. Its brain (Llama 3.3 70B) analyzes each message and picks the best AI from a pool of 5 models. Need a quick answer? The brain routes to the fast 8B model. Writing code? It picks Qwen or Llama 70B. Analyzing an image? Llama 4 Scout takes over.
It supports streaming (the response appears word by word), think mode (shows internal reasoning for hard tasks), web search, diagram generation, and image generation. The best option for most everyday uses.
Pollo Summum v1.0
| Brain model | Llama 3.3 70B |
|---|---|
| AI pool | Llama 8B, Llama 70B, Llama 4 Scout, Qwen 3 32B, GPT OSS 20B, GPT OSS 120B, Kimi K2 |
| Availability | Premium only |
| Max tokens | Up to 2,500 per response (depends on AI selected) |
| Streaming | Yes (progressive response) |
| Features | Chat, voice, images, code, chess, file analysis, chat memory |
| Best for | Deep analysis, complex tasks |
Pollo Summum is the most powerful tier, exclusive to premium users. It has access to all 7 AI models in the pool, including the heavyweight GPT OSS 120B and Kimi K2 — the most capable models for deep reasoning, science, and complex problem-solving.
Summum also unlocks exclusive features: file/document analysis (Pollo FILES integration) and chat memory (the AI remembers context from previous conversations). It reviews its own approach on every single message, ensuring maximum quality.
Tier Comparison
| Feature | Palos | Luces | Summum |
|---|---|---|---|
| Brain model | Llama 3.1 8B | Llama 3.3 70B | Llama 3.3 70B |
| AI pool size | 1 AI | 5 AIs | 7 AIs (all) |
| Max tokens/response | ~600 | Up to 2,000 | Up to 2,500 |
| Streaming | No | Yes | Yes |
| Speed | Very fast | Fast | Fast |
| Think mode | No | Yes | Yes |
| Web search | No | Yes | Yes |
| Image generation | No | Yes | Yes |
| Vision (image analysis) | No | Yes | Yes |
| Chess | No | Yes | Yes |
| File analysis | No | No | Yes |
| Chat memory | No | No | Yes |
| Availability | Free | Free | Premium |
Which one to choose
Palos if you need something fast and simple. Luces for daily use — it's the recommended option with access to 5 AIs and all core features. Summum when you need maximum depth and power, with access to all 7 AIs plus exclusive features like file analysis and chat memory (requires premium).
