""" Model registry for Noveum Trace SDK. Contains ModelInfo dataclass, MODEL_REGISTRY, and MODEL_ALIASES. Pricing is per 1M tokens in USD. Verify latest pricing from provider docs. Last updated: June 2026. """ from dataclasses import dataclass from typing import Optional @dataclass class ModelInfo: """Information about an LLM model.""" provider: str name: str context_window: int max_output_tokens: int input_cost_per_1m: float # Cost per 1M input tokens in USD output_cost_per_1m: float # Cost per 1M output tokens in USD supports_vision: bool = False supports_audio: bool = False supports_function_calling: bool = False training_cutoff: Optional[str] = None # --------------------------------------------------------------------------- # MODEL_REGISTRY # --------------------------------------------------------------------------- MODEL_REGISTRY: dict[str, ModelInfo] = { # ----------------------------------------------------------------------- # OpenAI GPT-4.1 Family # ----------------------------------------------------------------------- "gpt-4.1": ModelInfo( provider="openai", name="gpt-4.1", context_window=1047576, max_output_tokens=32768, input_cost_per_1m=3.00, output_cost_per_1m=12.00, supports_function_calling=True, training_cutoff="Jun 2024", ), "gpt-4.1-mini": ModelInfo( provider="openai", name="gpt-4.1-mini", context_window=1047576, max_output_tokens=32768, input_cost_per_1m=0.80, output_cost_per_1m=3.20, supports_function_calling=True, training_cutoff="Jun 2024", ), "gpt-4.1-nano": ModelInfo( provider="openai", name="gpt-4.1-nano", context_window=1047576, max_output_tokens=32768, input_cost_per_1m=0.20, output_cost_per_1m=0.80, supports_function_calling=True, training_cutoff="Jun 2024", ), # ----------------------------------------------------------------------- # OpenAI Reasoning Models # ----------------------------------------------------------------------- "o1": ModelInfo( provider="openai", name="o1", context_window=200000, max_output_tokens=100000, input_cost_per_1m=15.00, output_cost_per_1m=60.00, supports_function_calling=False, training_cutoff="Oct 2023", ), "o1-mini": ModelInfo( provider="openai", name="o1-mini", context_window=128000, max_output_tokens=65536, input_cost_per_1m=1.10, output_cost_per_1m=4.40, supports_function_calling=False, training_cutoff="Oct 2023", ), "o1-pro": ModelInfo( provider="openai", name="o1-pro", context_window=200000, max_output_tokens=100000, input_cost_per_1m=150.00, output_cost_per_1m=600.00, supports_function_calling=False, training_cutoff="Oct 2023", ), # ----------------------------------------------------------------------- # OpenAI GPT-4 Family # ----------------------------------------------------------------------- "gpt-4o": ModelInfo( provider="openai", name="gpt-4o", context_window=128000, max_output_tokens=16384, input_cost_per_1m=2.50, output_cost_per_1m=10.00, supports_vision=True, supports_function_calling=True, training_cutoff="Oct 2023", ), "gpt-4o-2024-11-20": ModelInfo( provider="openai", name="gpt-4o-2024-11-20", context_window=128000, max_output_tokens=16384, input_cost_per_1m=2.50, output_cost_per_1m=10.00, supports_vision=True, supports_function_calling=True, training_cutoff="Oct 2023", ), "gpt-4o-2024-08-06": ModelInfo( provider="openai", name="gpt-4o-2024-08-06", context_window=128000, max_output_tokens=16384, input_cost_per_1m=2.50, output_cost_per_1m=10.00, supports_vision=True, supports_function_calling=True, training_cutoff="Oct 2023", ), "gpt-4o-mini": ModelInfo( provider="openai", name="gpt-4o-mini", context_window=128000, max_output_tokens=16384, input_cost_per_1m=0.15, output_cost_per_1m=0.60, supports_vision=True, supports_function_calling=True, training_cutoff="Oct 2023", ), "gpt-4-turbo": ModelInfo( provider="openai", name="gpt-4-turbo", context_window=128000, max_output_tokens=4096, input_cost_per_1m=10.00, output_cost_per_1m=30.00, supports_vision=True, supports_function_calling=True, training_cutoff="Dec 2023", ), "gpt-4": ModelInfo( provider="openai", name="gpt-4", context_window=8192, max_output_tokens=8192, input_cost_per_1m=30.00, output_cost_per_1m=60.00, supports_function_calling=True, training_cutoff="Sep 2021", ), # ----------------------------------------------------------------------- # OpenAI GPT-5 Family # ----------------------------------------------------------------------- "gpt-5": ModelInfo( provider="openai", name="gpt-5", context_window=128000, max_output_tokens=16384, input_cost_per_1m=1.25, output_cost_per_1m=10.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "gpt-5-mini": ModelInfo( provider="openai", name="gpt-5-mini", context_window=128000, max_output_tokens=16384, input_cost_per_1m=0.25, output_cost_per_1m=2.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "gpt-5-nano": ModelInfo( provider="openai", name="gpt-5-nano", context_window=128000, max_output_tokens=16384, input_cost_per_1m=0.05, output_cost_per_1m=0.40, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "gpt-5-pro": ModelInfo( provider="openai", name="gpt-5-pro", context_window=128000, max_output_tokens=16384, input_cost_per_1m=15.00, output_cost_per_1m=120.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), # ----------------------------------------------------------------------- # OpenAI GPT-3.5 Family # ----------------------------------------------------------------------- "gpt-3.5-turbo": ModelInfo( provider="openai", name="gpt-3.5-turbo", context_window=16385, max_output_tokens=4096, input_cost_per_1m=0.50, output_cost_per_1m=1.50, supports_function_calling=True, training_cutoff="Sep 2021", ), "gpt-3.5-turbo-0125": ModelInfo( provider="openai", name="gpt-3.5-turbo-0125", context_window=16385, max_output_tokens=4096, input_cost_per_1m=0.50, output_cost_per_1m=1.50, supports_function_calling=True, training_cutoff="Sep 2021", ), # ----------------------------------------------------------------------- # Google Gemini Family # ----------------------------------------------------------------------- "gemini-2.5-flash": ModelInfo( provider="google", name="gemini-2.5-flash", context_window=1000000, max_output_tokens=8192, input_cost_per_1m=0.30, output_cost_per_1m=2.50, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "gemini-2.5-pro": ModelInfo( provider="google", name="gemini-2.5-pro", context_window=2000000, max_output_tokens=8192, input_cost_per_1m=1.25, output_cost_per_1m=10.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "gemini-2.0-flash": ModelInfo( provider="google", name="gemini-2.0-flash", context_window=1000000, max_output_tokens=8192, input_cost_per_1m=0.15, output_cost_per_1m=0.60, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), # Gemini 1.5 pricing published per 1K characters; converted to per 1M tokens # assuming 4 characters per token. "gemini-1.5-pro": ModelInfo( provider="google", name="gemini-1.5-pro", context_window=2000000, max_output_tokens=8192, input_cost_per_1m=1.25, output_cost_per_1m=5.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "gemini-1.5-flash": ModelInfo( provider="google", name="gemini-1.5-flash", context_window=1000000, max_output_tokens=8192, input_cost_per_1m=0.075, output_cost_per_1m=0.30, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), # ----------------------------------------------------------------------- # Anthropic Claude — Legacy (Claude 2 / Instant) # ----------------------------------------------------------------------- "claude-instant-1.2": ModelInfo( provider="anthropic", name="claude-instant-1.2", context_window=100000, max_output_tokens=4096, input_cost_per_1m=0.80, output_cost_per_1m=2.40, supports_function_calling=False, training_cutoff="Early 2023", ), "claude-2.0": ModelInfo( provider="anthropic", name="claude-2.0", context_window=100000, max_output_tokens=4096, input_cost_per_1m=8.00, output_cost_per_1m=24.00, supports_function_calling=False, training_cutoff="Early 2023", ), "claude-2.1": ModelInfo( provider="anthropic", name="claude-2.1", context_window=200000, max_output_tokens=4096, input_cost_per_1m=8.00, output_cost_per_1m=24.00, supports_function_calling=False, training_cutoff="Early 2023", ), # ----------------------------------------------------------------------- # Anthropic Claude 3 Family # ----------------------------------------------------------------------- "claude-3-haiku": ModelInfo( provider="anthropic", name="claude-3-haiku", context_window=200000, max_output_tokens=4096, input_cost_per_1m=0.25, output_cost_per_1m=1.25, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), "claude-3-haiku-20240307": ModelInfo( provider="anthropic", name="claude-3-haiku-20240307", context_window=200000, max_output_tokens=4096, input_cost_per_1m=0.25, output_cost_per_1m=1.25, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), "claude-3-sonnet": ModelInfo( provider="anthropic", name="claude-3-sonnet", context_window=200000, max_output_tokens=4096, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), "claude-3-sonnet-20240229": ModelInfo( provider="anthropic", name="claude-3-sonnet-20240229", context_window=200000, max_output_tokens=4096, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), "claude-3-opus": ModelInfo( provider="anthropic", name="claude-3-opus", context_window=200000, max_output_tokens=4096, input_cost_per_1m=15.00, output_cost_per_1m=75.00, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), "claude-3-opus-20240229": ModelInfo( provider="anthropic", name="claude-3-opus-20240229", context_window=200000, max_output_tokens=4096, input_cost_per_1m=15.00, output_cost_per_1m=75.00, supports_vision=True, supports_function_calling=True, training_cutoff="Aug 2023", ), # ----------------------------------------------------------------------- # Anthropic Claude 3.5 Family # ----------------------------------------------------------------------- "claude-3.5-haiku": ModelInfo( provider="anthropic", name="claude-3.5-haiku", context_window=200000, max_output_tokens=8192, input_cost_per_1m=0.80, output_cost_per_1m=4.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "claude-3.5-haiku-20241022": ModelInfo( provider="anthropic", name="claude-3.5-haiku-20241022", context_window=200000, max_output_tokens=8192, input_cost_per_1m=0.80, output_cost_per_1m=4.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "claude-3.5-sonnet": ModelInfo( provider="anthropic", name="claude-3.5-sonnet", context_window=200000, max_output_tokens=8192, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "claude-3.5-sonnet-20240620": ModelInfo( provider="anthropic", name="claude-3.5-sonnet-20240620", context_window=200000, max_output_tokens=8192, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "claude-3.5-sonnet-20241022": ModelInfo( provider="anthropic", name="claude-3.5-sonnet-20241022", context_window=200000, max_output_tokens=8192, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), # ----------------------------------------------------------------------- # Anthropic Claude 3.7 Family # ----------------------------------------------------------------------- "claude-3.7-sonnet": ModelInfo( provider="anthropic", name="claude-3.7-sonnet", context_window=200000, max_output_tokens=128000, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), "claude-3.7-sonnet-20250219": ModelInfo( provider="anthropic", name="claude-3.7-sonnet-20250219", context_window=200000, max_output_tokens=128000, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="Apr 2024", ), # ----------------------------------------------------------------------- # Anthropic Claude 4 Family # ----------------------------------------------------------------------- "claude-haiku-4": ModelInfo( provider="anthropic", name="claude-haiku-4", context_window=200000, max_output_tokens=8192, input_cost_per_1m=0.80, output_cost_per_1m=4.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "claude-sonnet-4": ModelInfo( provider="anthropic", name="claude-sonnet-4", context_window=200000, max_output_tokens=8192, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "claude-opus-4": ModelInfo( provider="anthropic", name="claude-opus-4", context_window=200000, max_output_tokens=8192, input_cost_per_1m=15.00, output_cost_per_1m=75.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), # Versioned Claude 4 IDs (date/patch suffixes) "claude-haiku-4-5": ModelInfo( provider="anthropic", name="claude-haiku-4-5", context_window=200000, max_output_tokens=8192, input_cost_per_1m=1.00, output_cost_per_1m=5.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "claude-sonnet-4-6": ModelInfo( provider="anthropic", name="claude-sonnet-4-6", context_window=200000, max_output_tokens=64000, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), "claude-opus-4-8": ModelInfo( provider="anthropic", name="claude-opus-4-8", context_window=200000, max_output_tokens=32768, input_cost_per_1m=15.00, output_cost_per_1m=75.00, supports_vision=True, supports_function_calling=True, training_cutoff="2025", ), # ----------------------------------------------------------------------- # Meta Llama Family # ----------------------------------------------------------------------- "llama-3.3-70b": ModelInfo( provider="meta", name="llama-3.3-70b", context_window=128000, max_output_tokens=2048, input_cost_per_1m=0.23, output_cost_per_1m=0.40, supports_function_calling=True, training_cutoff="Dec 2024", ), "llama-3.1-405b": ModelInfo( provider="meta", name="llama-3.1-405b", context_window=128000, max_output_tokens=2048, input_cost_per_1m=1.79, output_cost_per_1m=1.79, supports_function_calling=True, training_cutoff="Jul 2024", ), "llama-3.1-70b": ModelInfo( provider="meta", name="llama-3.1-70b", context_window=128000, max_output_tokens=2048, input_cost_per_1m=0.23, output_cost_per_1m=0.40, supports_function_calling=True, training_cutoff="Jul 2024", ), # ----------------------------------------------------------------------- # DeepSeek Models # ----------------------------------------------------------------------- "deepseek-v3": ModelInfo( provider="deepseek", name="deepseek-v3", context_window=128000, max_output_tokens=8192, input_cost_per_1m=0.14, output_cost_per_1m=0.28, supports_function_calling=True, training_cutoff="2024", ), "deepseek-r1": ModelInfo( provider="deepseek", name="deepseek-r1", context_window=128000, max_output_tokens=8192, input_cost_per_1m=0.55, output_cost_per_1m=2.19, supports_function_calling=True, training_cutoff="2024", ), # ----------------------------------------------------------------------- # Mistral Models # ----------------------------------------------------------------------- "mistral-large-2": ModelInfo( provider="mistral", name="mistral-large-2", context_window=128000, max_output_tokens=4096, input_cost_per_1m=2.00, output_cost_per_1m=6.00, supports_function_calling=True, training_cutoff="2024", ), "mistral-small-2409": ModelInfo( provider="mistral", name="mistral-small-2409", context_window=128000, max_output_tokens=4096, input_cost_per_1m=0.20, output_cost_per_1m=0.60, supports_function_calling=True, training_cutoff="2024", ), "mixtral-8x7b": ModelInfo( provider="mistral", name="mixtral-8x7b", context_window=32000, max_output_tokens=4096, input_cost_per_1m=0.50, output_cost_per_1m=0.50, supports_function_calling=True, training_cutoff="2024", ), # ----------------------------------------------------------------------- # Cohere Models # ----------------------------------------------------------------------- "command-r-plus": ModelInfo( provider="cohere", name="command-r-plus", context_window=128000, max_output_tokens=4096, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_function_calling=True, training_cutoff="2024", ), "command-r": ModelInfo( provider="cohere", name="command-r", context_window=128000, max_output_tokens=4096, input_cost_per_1m=0.50, output_cost_per_1m=1.50, supports_function_calling=True, training_cutoff="2024", ), # ----------------------------------------------------------------------- # OpenAI Embeddings — no output tokens; verify pricing before relying on # it for billing-critical enforcement, per this file's own pricing caveat. # ----------------------------------------------------------------------- "text-embedding-3-small": ModelInfo( provider="openai", name="text-embedding-3-small", context_window=8191, max_output_tokens=0, input_cost_per_1m=0.02, output_cost_per_1m=0.0, ), "text-embedding-3-large": ModelInfo( provider="openai", name="text-embedding-3-large", context_window=8191, max_output_tokens=0, input_cost_per_1m=0.13, output_cost_per_1m=0.0, ), "text-embedding-ada-002": ModelInfo( provider="openai", name="text-embedding-ada-002", context_window=8191, max_output_tokens=0, input_cost_per_1m=0.10, output_cost_per_1m=0.0, ), # ----------------------------------------------------------------------- # AWS Bedrock — keyed by the Bedrock model ID as returned by # normalize_model_name() (provider prefix like "anthropic."/"amazon." # is kept since it isn't a slash-separated prefix the normaliser strips). # Bedrock pricing can diverge from the vendor's direct API pricing for # the same underlying model, hence separate provider="bedrock" entries # rather than reusing the direct-API keys above. Verify latest pricing # from AWS Bedrock docs before relying on this for billing-critical # enforcement, per this file's own pricing caveat. # ----------------------------------------------------------------------- "anthropic.claude-3-5-sonnet": ModelInfo( provider="bedrock", name="anthropic.claude-3-5-sonnet", context_window=200000, max_output_tokens=8192, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_vision=True, supports_function_calling=True, ), "anthropic.claude-3-haiku": ModelInfo( provider="bedrock", name="anthropic.claude-3-haiku", context_window=200000, max_output_tokens=4096, input_cost_per_1m=0.25, output_cost_per_1m=1.25, supports_vision=True, supports_function_calling=True, ), "anthropic.claude-3-opus": ModelInfo( provider="bedrock", name="anthropic.claude-3-opus", context_window=200000, max_output_tokens=4096, input_cost_per_1m=15.00, output_cost_per_1m=75.00, supports_vision=True, supports_function_calling=True, ), "amazon.titan-text-express": ModelInfo( provider="bedrock", name="amazon.titan-text-express", context_window=8192, max_output_tokens=8192, input_cost_per_1m=0.20, output_cost_per_1m=0.60, ), "amazon.titan-text-lite": ModelInfo( provider="bedrock", name="amazon.titan-text-lite", context_window=4096, max_output_tokens=4096, input_cost_per_1m=0.15, output_cost_per_1m=0.20, ), "amazon.titan-embed-text": ModelInfo( provider="bedrock", name="amazon.titan-embed-text", context_window=8192, max_output_tokens=0, input_cost_per_1m=0.02, output_cost_per_1m=0.0, ), "meta.llama3-70b-instruct": ModelInfo( provider="bedrock", name="meta.llama3-70b-instruct", context_window=8192, max_output_tokens=2048, input_cost_per_1m=2.65, output_cost_per_1m=3.50, supports_function_calling=True, ), "meta.llama3-8b-instruct": ModelInfo( provider="bedrock", name="meta.llama3-8b-instruct", context_window=8192, max_output_tokens=2048, input_cost_per_1m=0.30, output_cost_per_1m=0.60, supports_function_calling=True, ), # Both the 2402 and 2407 Mistral Large snapshots ship on Bedrock and share # the same pricing, so key both — normalize_model_name() keeps the 4-digit # "-2402"/"-2407" (it only strips 8-digit dates), so one entry can't cover both. "mistral.mistral-large-2402": ModelInfo( provider="bedrock", name="mistral.mistral-large-2402", context_window=128000, max_output_tokens=8192, input_cost_per_1m=4.00, output_cost_per_1m=12.00, supports_function_calling=True, ), "mistral.mistral-large-2407": ModelInfo( provider="bedrock", name="mistral.mistral-large-2407", context_window=128000, max_output_tokens=8192, input_cost_per_1m=4.00, output_cost_per_1m=12.00, supports_function_calling=True, ), "cohere.command-r-plus": ModelInfo( provider="bedrock", name="cohere.command-r-plus", context_window=128000, max_output_tokens=4096, input_cost_per_1m=3.00, output_cost_per_1m=15.00, supports_function_calling=True, ), } # --------------------------------------------------------------------------- # MODEL_ALIASES — maps variant names → canonical registry keys # --------------------------------------------------------------------------- MODEL_ALIASES: dict[str, str] = { # OpenAI "gpt-4-turbo-preview": "gpt-4-turbo", "gpt-4-1106-preview": "gpt-4-turbo", "gpt-4-0125-preview": "gpt-4-turbo", "gpt-4-0613": "gpt-4", "gpt-4-0314": "gpt-4", "gpt-3.5-turbo-1106": "gpt-3.5-turbo", "gpt-3.5-turbo-16k": "gpt-3.5-turbo", # OpenAI GPT-5 shorthand "gpt5": "gpt-5", "gpt5-mini": "gpt-5-mini", "gpt5-nano": "gpt-5-nano", "gpt5-pro": "gpt-5-pro", # Common shorthand "gpt4": "gpt-4", "gpt4o": "gpt-4o", "gpt35": "gpt-3.5-turbo", # Anthropic — legacy versioned IDs already in registry as canonical entries; # these aliases cover alternate spellings / pre-normalisation names. "claude-3-opus-20240229": "claude-3-opus-20240229", # identity (normaliser strips date) "claude-3-sonnet-20240229": "claude-3-sonnet-20240229", "claude-3-haiku-20240307": "claude-3-haiku-20240307", "claude-3.5-sonnet-20241022": "claude-3.5-sonnet", "claude-3.5-haiku-20241022": "claude-3.5-haiku-20241022", # Alternate Claude 4 spellings "claude-4-opus": "claude-opus-4", "claude-4-sonnet": "claude-sonnet-4", "claude-4-haiku": "claude-haiku-4", # Shorthand "claude": "claude-3.5-sonnet", # Google "gemini-pro": "gemini-1.5-pro", "gemini-flash": "gemini-1.5-flash", "gemini": "gemini-1.5-pro", "text-embedding-004": "gemini-text-embedding", }