Skip to main content
Polpo maintains a live model catalog sourced from the @mariozechner/pi-ai library. Every model in the catalog includes structured metadata: cost per token, context window size, max output tokens, supported input types, and whether it supports reasoning mode.

​
Browsing the Catalog

​
Programmatic Access

The catalog is accessible through the Polpo API: 
import { listProviders, listModels, getModelInfo } from "polpo/llm";

// List all available providers
const providers = listProviders();
// → ["anthropic", "openai", "google", "groq", "mistral", ...]

// List all models for a provider
const anthropicModels = listModels("anthropic");
// → [{ id: "claude-opus-4-6", name: "Claude Opus", provider: "anthropic", ... }, ...]

// List ALL models across ALL providers
const allModels = listModels();
// → hundreds of models

// Get details for a specific model
const info = getModelInfo("anthropic:claude-sonnet-4-5-20250929");
// → { id: "claude-sonnet-4-5-20250929", reasoning: true, contextWindow: 200000, ... }

​
In System Prompts

Polpo dynamically generates model listings for system prompts using buildModelListingForPrompt(). This means agents always have an up-to-date view of available models — no hardcoded lists.

​
Model Metadata

Every model in the catalog carries these fields:
FieldTypeDescription
idstringModel ID used in API calls (e.g. claude-sonnet-4-5-20250929)
namestringHuman-readable name (e.g. Claude Sonnet 4.5)
providerstringProvider ID (e.g. anthropic)
reasoningbooleanSupports extended thinking / chain-of-thought
inputstring[]Supported input types: ["text"], ["text", "image"]
contextWindownumberMax input tokens (e.g. 200000)
maxTokensnumberMax output tokens (e.g. 8192)
cost.inputnumberCost per 1M input tokens (USD)
cost.outputnumberCost per 1M output tokens (USD)
cost.cacheReadnumberCost per 1M cache-read tokens (USD)
cost.cacheWritenumberCost per 1M cache-write tokens (USD)

​
By Use Case

Use CaseRecommended ModelWhy
Best all-roundanthropic:claude-sonnet-4-6Latest Claude, strong coding and reasoning
Most capableanthropic:claude-opus-4-6Complex architecture, multi-step reasoning
Fast + cheapanthropic:claude-haiku-4-5-20251001Quick tasks, boilerplate generation
Deep reasoningopenai:o3Code review, complex logic problems
Latest GPTopenai:gpt-5.2Newest OpenAI flagship with reasoning
Fast inferencegroq:llama-3.3-70b-versatileHigh throughput, lower quality ceiling
Long contextgoogle:gemini-2.5-proLarge codebases, 1M+ token context
Latest Geminigoogle:gemini-3-pro-previewState-of-art reasoning and agentic coding
Code generationmistral:codestral-latestSpecialized code model
Budget reasoningopenai:o4-miniReasoning at lower cost
Self-hostedollama:qwen2.5-coder:32bZero API cost, full privacy

​
By Agent Role

A recommended multi-agent team configuration: 
{
  "agents": [
    {
      "name": "architect",
      "model": "anthropic:claude-opus-4-6",
      "role": "System design — needs deep reasoning and broad context"
    },
    {
      "name": "coder",
      "model": "anthropic:claude-sonnet-4-6",
      "role": "Implementation — best balance of speed and quality"
    },
    {
      "name": "reviewer",
      "model": "openai:o3",
      "role": "Code review — deep reasoning catches subtle bugs"
    },
    {
      "name": "fast-worker",
      "model": "groq:llama-3.3-70b-versatile",
      "role": "Boilerplate, formatting, simple refactoring — speed matters"
    },
    {
      "name": "researcher",
      "model": "google:gemini-2.5-pro",
      "role": "Large codebase analysis — needs massive context window"
    }
  ],
  "settings": {
    "orchestratorModel": "anthropic:claude-sonnet-4-6"
  }
}

​
Cost Tiers

TierModelsInput $/1MOutput $/1MBest For
Freeopencode:big-pickle$0$0Testing, experiments
Budgetclaude-haiku-4-5, gpt-4o-mini, gemini-2.5-flash, o4-mini0.15−0.15 - 10.60−0.60 - 5High-volume simple tasks
Standardclaude-sonnet-4-6, gpt-5.2, gpt-4o, gemini-2.5-pro1.75−1.75 - 58−8 - 15General coding, daily work
Premiumclaude-opus-4-6, o3, grok-43−3 - 1515−15 - 75Complex architecture, review
LocalOllama, vLLM, LM Studio$0$0Privacy, offline, custom models

​
Model Capabilities

​
Reasoning Models

Models with reasoning: true support extended thinking, where the model generates a chain-of-thought before producing the final answer. These are best for complex tasks:
  • anthropic:claude-opus-4-6, claude-sonnet-4-6, claude-sonnet-4-5 (extended thinking)
  • openai:o3, o3-pro, o4-mini (chain-of-thought)
  • openai:gpt-5.2, gpt-5.1, gpt-5 (built-in reasoning)
  • google:gemini-3-pro-preview, gemini-2.5-pro, gemini-2.5-flash (thinking)
  • xai:grok-4, grok-3-mini, grok-code-fast-1 (reasoning)
  • mistral:magistral-medium-latest, magistral-small-latest (chain-of-thought)
  • groq:deepseek-r1-distill-llama-70b, qwen/qwen3-32b (open reasoning)

​
Vision Models

Models with input: ["text", "image"] can process images:
  • anthropic:claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5
  • openai:gpt-4o, gpt-4.1, gpt-5.2
  • google:gemini-3-pro-preview, gemini-2.5-pro, gemini-2.5-flash
  • xai:grok-4, grok-4-1-fast-reasoning
  • mistral:mistral-large-latest, pixtral-large-latest
Use the imageModel setting to designate a vision-capable model for image analysis tasks: 
{
  "settings": {
    "imageModel": "anthropic:claude-sonnet-4-5-20250929"
  }
}

​
Context Window Comparison

ModelContext WindowNotes
xai:grok-42,000,000Largest available
google:gemini-2.5-pro1,048,576Largest Gemini
google:gemini-3-pro-preview1,048,576Latest Gemini
openai:gpt-4.11,000,000Largest GPT-4
mistral:codestral-latest256,000Code-specialized
anthropic:claude-sonnet-4-6200,000Standard for Claude
anthropic:claude-opus-4-6200,000Standard for Claude
openai:o3200,000Reasoning
openai:gpt-4o128,000Standard GPT
groq:llama-3.3-70b-versatile131,072Open model

​
Model Allowlist

For enterprise environments, you can restrict which models agents are allowed to use: 
{
  "settings": {
    "modelAllowlist": {
      "anthropic:claude-sonnet-4-5-20250929": {
        "alias": "Sonnet",
        "params": {}
      },
      "anthropic:claude-haiku-4-5-20251001": {
        "alias": "Haiku",
        "params": {}
      },
      "openai:gpt-4o": {
        "alias": "GPT",
        "params": {}
      }
    }
  }
}
When a modelAllowlist is configured, any agent or plan that tries to use a model not in the list will be rejected. The /model in-chat command also enforces the allowlist. This is useful for:
  • Cost control — only allow budget-tier models
  • Compliance — restrict to approved vendors
  • Consistency — ensure all agents use tested models

​
CLI: polpo models

​
polpo models list

Browse the full catalog from the command line: 
# Featured models by provider
polpo models list

# All models for a provider
polpo models list --provider anthropic

# All models (no truncation)
polpo models list --all

# Machine-readable formats
polpo models list --plain    # one per line: provider:model
polpo models list --json     # full JSON with metadata

​
polpo models status

Show your resolved model configuration, fallback chain, and auth status: 
polpo models status

Model Configuration

  * Primary: anthropic:claude-sonnet-4-5-20250929
    Claude Sonnet 4.5 | 200k ctx | reasoning

  Fallback chain:
    1. openai:gpt-4o
    2. groq:llama-3.3-70b-versatile

  Provider Auth:

    anthropic: env + oauth (2 profiles)
    openai: env
    groq: env
Use --check in CI to verify auth is working (exit code 1 = missing, 2 = expiring): 
polpo models status --check

​
polpo models scan

Discover locally running model servers: 
polpo models scan
Probes standard ports for Ollama (11434), vLLM (8000), LM Studio (1234), LocalAI (8080), LiteLLM (4000), and TGI (8080). Reports discovered servers and their loaded models.

​
In-Chat: /model Command

During a chat session, switch models without restarting:
CommandAction
/modelShow available models as a numbered list
/model 3Select model #3 from the list
/model anthropic/claude-opus-4-6Switch to a specific model
/model statusShow current model and auth status
/model resetReturn to the default model
You can also pin a specific auth profile for the session: 
/model claude-opus-4-6@anthropic:user@example.com
See the Authentication guide for full /model syntax.