llm_models_spider

Auto-updated registry of LLM model capabilities, arena rankings, and pricing.

Features

• Zero runtime dependencies - Pure compile-time lookups
• Multi-source data - Aggregated from OpenRouter, LiteLLM, and Chatbot Arena
• Rich model profiles - Capabilities, pricing, context windows, and arena rankings
• Auto-updated - Model lists fetched daily via GitHub Actions
• Fallback patterns - Unknown models matched against common naming patterns

Installation

[dependencies]
llm_models_spider = "0.1"

Usage

Capability Checks

use llm_models_spider::{supports_vision, supports_audio, supports_video, supports_pdf, is_text_only};

// Vision support
assert!(supports_vision("gpt-4o"));
assert!(supports_vision("claude-3-sonnet-20240229"));
assert!(supports_vision("google/gemini-2.0-flash"));
assert!(!supports_vision("gpt-3.5-turbo"));

// Audio, video, PDF support
assert!(supports_audio("gemini-2.0-flash"));
assert!(supports_video("gemini-2.0-flash"));
assert!(supports_pdf("claude-3-sonnet"));

// Text-only check
assert!(is_text_only("gpt-3.5-turbo"));
assert!(is_text_only("llama-3-70b"));

Model Profiles

Get full model intelligence — capabilities, pricing, context window, and arena ranking:

use llm_models_spider::{model_profile, arena_rank};

if let Some(profile) = model_profile("gpt-4o") {
    println!("Vision: {}", profile.capabilities.vision);
    println!("Max input: {} tokens", profile.max_input_tokens);
    println!("Max output: {} tokens", profile.max_output_tokens);

    if let Some(cost) = profile.pricing.input_cost_per_m_tokens {
        println!("Input cost: ${:.2}/M tokens", cost);
    }
    if let Some(cost) = profile.pricing.output_cost_per_m_tokens {
        println!("Output cost: ${:.2}/M tokens", cost);
    }
}

// Arena ranking (0.0-100.0, higher is better)
if let Some(rank) = arena_rank("gpt-4o") {
    println!("Arena rank: {:.1}/100", rank);
}

Full Capabilities Lookup

use llm_models_spider::ModelCapabilities;

if let Some(caps) = ModelCapabilities::lookup("gemini-2.0-flash") {
    println!("Vision: {}", caps.vision);
    println!("Audio: {}", caps.audio);
    println!("Video: {}", caps.video);
    println!("File/PDF: {}", caps.file);
}

Direct Model Info Access

For advanced use cases, access the raw MODEL_INFO array directly:

use llm_models_spider::MODEL_INFO;

// MODEL_INFO is sorted by name — use binary search for fast lookups
let idx = MODEL_INFO.binary_search_by(|e| e.name.cmp("gpt-4o"));
if let Ok(idx) = idx {
    let info = &MODEL_INFO[idx];
    println!("{}: vision={}, audio={}, arena={}",
        info.name, info.supports_vision, info.supports_audio, info.arena_overall);
}

Data Sources

Model data is aggregated from three sources:

Source	Data Provided
OpenRouter API	Modalities (vision, audio), pricing, context length
LiteLLM	Vision, audio, video, PDF support, pricing, context window
Chatbot Arena	Arena Elo scores (normalized to 0-100)

Merge strategy:

• Capabilities: OR across sources (if any source says vision=true, it's vision)
• Pricing: Prefer LiteLLM (more granular), fall back to OpenRouter
• Context window: Take max across sources
• Arena scores: From Chatbot Arena only

Fallback Patterns

For models not yet in any data source, the library falls back to pattern matching:

Pattern	Examples
gpt-4o, gpt-4-turbo, gpt-4-vision	OpenAI vision models
claude-3, claude-4	Anthropic multimodal
gemini-1.5, gemini-2	Google Gemini
qwen2-vl, qwen-vl	Alibaba Qwen VL
-vision, -vl suffix	Generic vision indicators

Manual Update

To manually update the model lists:

cargo run --bin update-models --features updater

License

MIT License - see LICENSE for details.