LLMSpecs

LLMSpecs is a lightweight Ruby interface for fetching large language model (LLM) specifications from the Parsera API. It provides a simple, efficient way to access model metadata with built-in caching and query support.

Features

Fetches LLM specs from: https://api.parsera.org/v1/llm-specs
Caches results locally to llm-specs.json
Provides a queryable catalog of models via .models
Supports filtering and lookup by model attributes

Example Usage

require 'llm_specs'

models = LLMSpecs.models # all models

models = LLMSpecs.gemini # provider specific models
models = LLMSpecs.anthropic
models = LLMSpecs.openai
models = LLMSpecs.deepseek

providers = LLMSpecs.providers # a list of all providers covered by Parsera API

Find a model by ID

model = LLMSpecs.models.find("claude-3-sonnet")

Filter models

# List all Anthropic models
anthropic_models = LLMSpecs.models.where(provider: "anthropic")
# or 
anthropic_models = LLMSpecs.anthropic

# List all Anthropic models by family "claude-3-7-sonnet"
claude_models = LLMSpecs.models.where(provider: "anthropic", family: "claude-3-7-sonnet")
#or 
claude_models = LLMSpecs.anthropic.where(family: "claude-3-7-sonnet")

# List all models that support streaming
streaming_models = LLMSpecs.models.select(&:streaming?)

# List all models that support audio output
audio_models = LLMSpecs.models.select(&:audio_output?)

Model Capabilities

Each model is represented by an instance of LLMSpecs::Model, a value object that encapsulates:

id, name, provider, family
Token limits (context_window, max_output_tokens)
Supported modalities (e.g. text, image)
Capabilities such as:
- function_calling?
- structured_output?
- streaming?
- reasoning?
- citations?
- batch?
Pricing breakdowns via pricing, input_pricing and output_pricing

You can easily check capabilities:

model.supports?(:function_calling) # => true/false
model.function_calling?  # => shortcut for above
# with multiple arguments
model.supports?(:function_calling, :batch) # => true/false

or input or output modalities:

model.supports_input?(:audio) # => true/false
model.supports_input?(:audio, :text) # multiple argument
model.audio_input? # shortcut
model.image_input?
model.text_input?

model.supports_output?(:audio) # => true/false
model.supports_output?(:audio, :embeddings) # multiple argument
model.audio_output? # shortcut
model.image_output?
model.text_output?
model.embeddings_output?

Pricing methods:

model.pricing # => returns a hash: 
#  {
#    text_tokens: {
#      standard: {
#        input_per_million: 15.0,
#        cached_input_per_million: 18.75,
#        output_per_million: 75.0
#      },
#      batch: {
#        input_per_million: nil,
#        output_per_million: nil
#      }
#    },
#    embeddings: {
#      standard: {
#        input_per_million: nil
#      },
#      batch: {
#        input_per_million: nil
#      }
#    }
#  }

model.pricing.dig(:text_tokens, :standard, :input_per_million)
# input/output pricing
model.input_pricing  # 15.0 => $ per 1M input tokens (default to pricing[:text_tokens][:standard][:input_per_million])
model.output_pricing # 75.0 => $ per 1M output tokens

model.input_pricing(:text_tokens, :batch)
model.input_pricing(:embeddings)

Cache configuration

LLMSpecs.cache_path = Rails.root.join("tmp", "cache", "llm-specs-cache.json")

License

The gem is available as open source under the terms of the MIT License.

llm_specs

Development