diff --git a/litellm-rust/crates/model-catalog/src/model_info.rs b/litellm-rust/crates/model-catalog/src/model_info.rs index 96dec84de84..e356277ea08 100644 --- a/litellm-rust/crates/model-catalog/src/model_info.rs +++ b/litellm-rust/crates/model-catalog/src/model_info.rs @@ -28,6 +28,12 @@ pub struct ModelInfo { /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] pub cache_creation_input_token_cost_above_32k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub cache_creation_input_token_cost_above_100k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub cache_creation_input_token_cost_above_100k_tokens_batches: Option, #[serde(skip_serializing_if = "Option::is_none")] pub cache_creation_input_token_cost_above_128k_tokens: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. @@ -35,6 +41,9 @@ pub struct ModelInfo { pub cache_creation_input_token_cost_above_1hr: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] + pub cache_creation_input_token_cost_above_1hr_above_100k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] pub cache_creation_input_token_cost_above_1hr_above_200k_tokens: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] @@ -81,6 +90,12 @@ pub struct ModelInfo { /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] pub cache_read_input_token_cost_above_32k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub cache_read_input_token_cost_above_100k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub cache_read_input_token_cost_above_100k_tokens_batches: Option, #[serde(skip_serializing_if = "Option::is_none")] pub cache_read_input_token_cost_above_128k_tokens: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. @@ -199,6 +214,12 @@ pub struct ModelInfo { /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] pub input_cost_per_token_above_32k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub input_cost_per_token_above_100k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub input_cost_per_token_above_100k_tokens_batches: Option, #[serde(skip_serializing_if = "Option::is_none")] pub input_cost_per_token_above_128k_tokens: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. @@ -354,6 +375,12 @@ pub struct ModelInfo { /// Rate applied once the prompt exceeds the token threshold in the field name. #[serde(skip_serializing_if = "Option::is_none")] pub output_cost_per_token_above_32k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub output_cost_per_token_above_100k_tokens: Option, + /// Rate applied once the prompt exceeds the token threshold in the field name. + #[serde(skip_serializing_if = "Option::is_none")] + pub output_cost_per_token_above_100k_tokens_batches: Option, #[serde(skip_serializing_if = "Option::is_none")] pub output_cost_per_token_above_128k_tokens: Option, /// Rate applied once the prompt exceeds the token threshold in the field name. diff --git a/litellm-rust/crates/model-catalog/tests/registry_validation.rs b/litellm-rust/crates/model-catalog/tests/registry_validation.rs index 8f555e1f884..67589010aba 100644 --- a/litellm-rust/crates/model-catalog/tests/registry_validation.rs +++ b/litellm-rust/crates/model-catalog/tests/registry_validation.rs @@ -74,3 +74,20 @@ fn checked_in_catalog_and_backup_match() { "invalid registry aliases" ); } + +#[rstest] +#[case::input("input_cost_per_token_above_100k_tokens")] +#[case::input_batches("input_cost_per_token_above_100k_tokens_batches")] +#[case::output("output_cost_per_token_above_100k_tokens")] +#[case::output_batches("output_cost_per_token_above_100k_tokens_batches")] +#[case::cache_creation("cache_creation_input_token_cost_above_100k_tokens")] +#[case::cache_creation_batches("cache_creation_input_token_cost_above_100k_tokens_batches")] +#[case::cache_creation_1hr("cache_creation_input_token_cost_above_1hr_above_100k_tokens")] +#[case::cache_read("cache_read_input_token_cost_above_100k_tokens")] +#[case::cache_read_batches("cache_read_input_token_cost_above_100k_tokens_batches")] +fn registry_validation_keeps_above_100k_tier_rates(#[case] field: &str) { + let mut entry = Map::new(); + entry.insert("litellm_provider".into(), "anthropic".into()); + entry.insert(field.into(), 5e-7.into()); + validate_model_entry("test", &Value::Object(entry)).unwrap(); +} diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index ef601bf522f..274ed928073 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -80770,6 +80770,10 @@ "supports_vision": true }, "claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens_batches": 2.5e-07, + "output_cost_per_token_above_100k_tokens_batches": 1.25e-06, + "cache_creation_input_token_cost_above_100k_tokens_batches": 3.125e-07, + "cache_read_input_token_cost_above_100k_tokens_batches": 2.5e-08, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -80821,6 +80825,11 @@ "cache_read_input_token_cost_above_100k_tokens": 5e-08 }, "bedrock_mantle/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -80862,6 +80871,11 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-haiku-5-5.html" }, "bedrock_mantle/us-gov-west-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -80898,6 +80912,11 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-haiku-5-5.html" }, "anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -80939,6 +80958,11 @@ "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, "apac.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "litellm_provider": "bedrock_converse", "supports_tool_search": true, @@ -80975,6 +80999,11 @@ "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, "au.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81016,6 +81045,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "azure_ai/claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "supports_mid_conversation_system": true, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -81052,6 +81086,11 @@ "source": "https://management.azure.com/subscriptions/c873328e-b572-4770-8dff-aaeb6f1f0e79/providers/Microsoft.CognitiveServices/locations/eastus2/models?api-version=2024-10-01" }, "bedrock/us-gov-east-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81087,6 +81126,11 @@ "supports_xhigh_reasoning_effort": true }, "bedrock/us-gov-west-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81122,6 +81166,11 @@ "supports_xhigh_reasoning_effort": true }, "eu.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81163,6 +81212,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "global.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -81204,6 +81258,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "jp.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81256,6 +81315,11 @@ "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "us-gov.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81292,6 +81356,11 @@ "supports_xhigh_reasoning_effort": true }, "us.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index ef601bf522f..274ed928073 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -80770,6 +80770,10 @@ "supports_vision": true }, "claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens_batches": 2.5e-07, + "output_cost_per_token_above_100k_tokens_batches": 1.25e-06, + "cache_creation_input_token_cost_above_100k_tokens_batches": 3.125e-07, + "cache_read_input_token_cost_above_100k_tokens_batches": 2.5e-08, "supports_anthropic_compaction": true, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -80821,6 +80825,11 @@ "cache_read_input_token_cost_above_100k_tokens": 5e-08 }, "bedrock_mantle/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -80862,6 +80871,11 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-haiku-5-5.html" }, "bedrock_mantle/us-gov-west-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -80898,6 +80912,11 @@ "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-haiku-5-5.html" }, "anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -80939,6 +80958,11 @@ "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, "apac.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "litellm_provider": "bedrock_converse", "supports_tool_search": true, @@ -80975,6 +80999,11 @@ "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json" }, "au.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81016,6 +81045,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "azure_ai/claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "supports_mid_conversation_system": true, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -81052,6 +81086,11 @@ "source": "https://management.azure.com/subscriptions/c873328e-b572-4770-8dff-aaeb6f1f0e79/providers/Microsoft.CognitiveServices/locations/eastus2/models?api-version=2024-10-01" }, "bedrock/us-gov-east-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81087,6 +81126,11 @@ "supports_xhigh_reasoning_effort": true }, "bedrock/us-gov-west-1/anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81122,6 +81166,11 @@ "supports_xhigh_reasoning_effort": true }, "eu.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81163,6 +81212,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "global.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.25e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.25e-07, "cache_creation_input_token_cost_above_1hr": 2e-07, @@ -81204,6 +81258,11 @@ "source": "https://aws.amazon.com/bedrock/pricing/" }, "jp.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, @@ -81256,6 +81315,11 @@ "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "us-gov.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 6e-07, + "output_cost_per_token_above_100k_tokens": 3e-06, + "cache_creation_input_token_cost_above_100k_tokens": 7.5e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.2e-06, + "cache_read_input_token_cost_above_100k_tokens": 6e-08, "bedrock_converse_supports_strict_tools": false, "bedrock_output_config_effort_ceiling": "xhigh", "cache_creation_input_token_cost": 1.5e-07, @@ -81292,6 +81356,11 @@ "supports_xhigh_reasoning_effort": true }, "us.anthropic.claude-haiku-5-5": { + "input_cost_per_token_above_100k_tokens": 5.5e-07, + "output_cost_per_token_above_100k_tokens": 2.75e-06, + "cache_creation_input_token_cost_above_100k_tokens": 6.875e-07, + "cache_creation_input_token_cost_above_1hr_above_100k_tokens": 1.1e-06, + "cache_read_input_token_cost_above_100k_tokens": 5.5e-08, "bedrock_converse_supports_strict_tools": false, "cache_creation_input_token_cost": 1.375e-07, "cache_creation_input_token_cost_above_1hr": 2.2e-07, diff --git a/model_prices_and_context_window.schema.json b/model_prices_and_context_window.schema.json index 41b616a6b1f..e51ae0453ea 100644 --- a/model_prices_and_context_window.schema.json +++ b/model_prices_and_context_window.schema.json @@ -88,6 +88,11 @@ "minimum": 0, "description": "Rate applied once the prompt exceeds the token threshold in the field name." }, + "cache_creation_input_token_cost_above_100k_tokens_batches": { + "type": "number", + "minimum": 0, + "description": "Rate applied once the prompt exceeds the token threshold in the field name." + }, "cache_creation_input_token_cost_above_128k_tokens": { "type": "number", "minimum": 0, @@ -189,6 +194,11 @@ "minimum": 0, "description": "Rate applied once the prompt exceeds the token threshold in the field name." }, + "cache_read_input_token_cost_above_100k_tokens_batches": { + "type": "number", + "minimum": 0, + "description": "Rate applied once the prompt exceeds the token threshold in the field name." + }, "cache_read_input_token_cost_above_128k_tokens": { "type": "number", "minimum": 0, @@ -397,6 +407,11 @@ "minimum": 0, "description": "Rate applied once the prompt exceeds the token threshold in the field name." }, + "input_cost_per_token_above_100k_tokens_batches": { + "type": "number", + "minimum": 0, + "description": "Rate applied once the prompt exceeds the token threshold in the field name." + }, "input_cost_per_token_above_128k_tokens": { "type": "number", "minimum": 0, @@ -781,6 +796,11 @@ "minimum": 0, "description": "Rate applied once the prompt exceeds the token threshold in the field name." }, + "output_cost_per_token_above_100k_tokens_batches": { + "type": "number", + "minimum": 0, + "description": "Rate applied once the prompt exceeds the token threshold in the field name." + }, "output_cost_per_token_above_128k_tokens": { "type": "number", "minimum": 0, diff --git a/tests/unit/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py b/tests/unit/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py index ea628794c2c..9c5714eeb11 100644 --- a/tests/unit/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py +++ b/tests/unit/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py @@ -3830,3 +3830,61 @@ def test_azure_gpt_5_6_alias_matches_sol_pricing(_local_model_cost_map, region_p assert shared_cost_fields for field in shared_cost_fields: assert alias[field] == sol[field], field + + +@pytest.mark.parametrize( + ("model", "custom_llm_provider", "prompt_tokens", "input_rate", "cache_read_rate", "output_rate"), + [ + ("claude-haiku-5-5", "anthropic", 100_000, 1e-07, 1e-08, 5e-07), + ("claude-haiku-5-5", "anthropic", 100_001, 5e-07, 5e-08, 2.5e-06), + ("azure_ai/claude-haiku-5-5", "azure_ai", 100_000, 1e-07, 1e-08, 5e-07), + ("azure_ai/claude-haiku-5-5", "azure_ai", 100_001, 5e-07, 5e-08, 2.5e-06), + ("global.anthropic.claude-haiku-5-5", "bedrock", 100_000, 1e-07, 1e-08, 5e-07), + ("global.anthropic.claude-haiku-5-5", "bedrock", 100_001, 5e-07, 5e-08, 2.5e-06), + ("us.anthropic.claude-haiku-5-5", "bedrock", 100_000, 1.1e-07, 1.1e-08, 5.5e-07), + ("us.anthropic.claude-haiku-5-5", "bedrock", 100_001, 5.5e-07, 5.5e-08, 2.75e-06), + ("us-gov.anthropic.claude-haiku-5-5", "bedrock", 100_000, 1.2e-07, 1.2e-08, 6e-07), + ("us-gov.anthropic.claude-haiku-5-5", "bedrock", 100_001, 6e-07, 6e-08, 3e-06), + ("bedrock_mantle/anthropic.claude-haiku-5-5", "bedrock_mantle", 100_001, 5.5e-07, 5.5e-08, 2.75e-06), + ("bedrock/us-gov-west-1/anthropic.claude-haiku-5-5", "bedrock", 100_001, 6e-07, 6e-08, 3e-06), + ], +) +def test_generic_cost_per_token_claude_haiku_5_5_prompt_length_tiers( + _local_model_cost_map, model, custom_llm_provider, prompt_tokens, input_rate, cache_read_rate, output_rate +): + """Claude Haiku 5.5 bills every token at 5x the base rates once the prompt is over 100,000 tokens.""" + cached_tokens = 10_000 + completion_tokens = 1_000 + usage = Usage( + prompt_tokens=prompt_tokens, + completion_tokens=completion_tokens, + total_tokens=prompt_tokens + completion_tokens, + prompt_tokens_details=PromptTokensDetailsWrapper(cached_tokens=cached_tokens), + ) + prompt_cost, completion_cost = generic_cost_per_token( + model=model, + usage=usage, + custom_llm_provider=custom_llm_provider, + ) + expected_prompt = (prompt_tokens - cached_tokens) * input_rate + cached_tokens * cache_read_rate + assert prompt_cost == pytest.approx(expected_prompt) + assert completion_cost == pytest.approx(completion_tokens * output_rate) + + +@pytest.mark.parametrize( + ("prompt_tokens", "expected"), + [ + (100_000, (5e-08, 2.5e-07, 5e-09, 6.25e-08)), + (100_001, (2.5e-07, 1.25e-06, 2.5e-08, 3.125e-07)), + ], +) +def test_get_batch_cost_rates_claude_haiku_5_5_prompt_length_tiers(_local_model_cost_map, prompt_tokens, expected): + from litellm.litellm_core_utils.llm_cost_calc.utils import get_batch_cost_rates + + rates = get_batch_cost_rates( + litellm.model_cost["claude-haiku-5-5"], + Usage(prompt_tokens=prompt_tokens, completion_tokens=1, total_tokens=prompt_tokens + 1), + "anthropic", + ) + + assert (rates.input, rates.output, rates.cache_read, rates.cache_creation) == expected diff --git a/tests/unit/test_utils.py b/tests/unit/test_utils.py index 1f420d0e8dc..852da82e368 100644 --- a/tests/unit/test_utils.py +++ b/tests/unit/test_utils.py @@ -778,6 +778,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "cache_creation_input_token_cost_above_1hr": {"type": "number"}, "cache_creation_input_token_cost_above_32k_tokens": {"type": "number"}, "cache_creation_input_token_cost_above_100k_tokens": {"type": "number"}, + "cache_creation_input_token_cost_above_100k_tokens_batches": {"type": "number"}, "cache_creation_input_token_cost_above_128k_tokens": {"type": "number"}, "cache_creation_input_token_cost_above_200k_tokens": {"type": "number"}, "cache_creation_input_token_cost_above_256k_tokens": {"type": "number"}, @@ -794,6 +795,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "cache_read_input_token_cost": {"type": "number"}, "cache_read_input_token_cost_above_32k_tokens": {"type": "number"}, "cache_read_input_token_cost_above_100k_tokens": {"type": "number"}, + "cache_read_input_token_cost_above_100k_tokens_batches": {"type": "number"}, "cache_read_input_token_cost_above_128k_tokens": {"type": "number"}, "cache_read_input_token_cost_above_200k_tokens": {"type": "number"}, "cache_read_input_token_cost_above_200k_tokens_batches": {"type": "number"}, @@ -823,6 +825,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "input_cost_per_video_token": {"type": "number"}, "input_cost_per_token_above_32k_tokens": {"type": "number"}, "input_cost_per_token_above_100k_tokens": {"type": "number"}, + "input_cost_per_token_above_100k_tokens_batches": {"type": "number"}, "input_cost_per_token_above_200k_tokens": {"type": "number"}, "input_cost_per_token_above_200k_tokens_batches": {"type": "number"}, "input_cost_per_token_above_256k_tokens": {"type": "number"}, @@ -933,6 +936,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "output_cost_per_token": {"type": "number"}, "output_cost_per_token_above_32k_tokens": {"type": "number"}, "output_cost_per_token_above_100k_tokens": {"type": "number"}, + "output_cost_per_token_above_100k_tokens_batches": {"type": "number"}, "output_cost_per_token_above_128k_tokens": {"type": "number"}, "output_cost_per_token_above_200k_tokens": {"type": "number"}, "output_cost_per_token_above_200k_tokens_batches": {"type": "number"},