diff --git a/tests/integration/cost_calculation/cost_map.py b/tests/integration/cost_calculation/cost_map.py index 8f11b3ce2f1..ffdb6752bf2 100644 --- a/tests/integration/cost_calculation/cost_map.py +++ b/tests/integration/cost_calculation/cost_map.py @@ -1603,32 +1603,44 @@ COST_MAP_ENTRIES: Final[dict[str, dict[str, JsonValue]]] = { }, "azure/gpt-6-luna": { "cache_read_input_token_cost": 1.3e-08, + "cache_read_input_token_cost_flex": 6.5e-09, "cache_creation_input_token_cost": 1.625e-07, + "cache_creation_input_token_cost_flex": 8.125e-08, "cache_creation_input_token_cost_above_272k_tokens": 3.25e-07, + "cache_creation_input_token_cost_above_272k_tokens_flex": 1.625e-07, "input_cost_per_token": 1.3e-07, + "input_cost_per_token_flex": 6.5e-08, "litellm_provider": "azure", "max_input_tokens": 922000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 6.5e-07, + "output_cost_per_token_flex": 3.25e-07, "supports_function_calling": True, "supports_none_reasoning_effort": True, "input_cost_per_token_above_272k_tokens": 2.6e-07, + "input_cost_per_token_above_272k_tokens_flex": 1.3e-07, "output_cost_per_token_above_272k_tokens": 1.3e-06, + "output_cost_per_token_above_272k_tokens_flex": 4.875e-07, "cache_read_input_token_cost_above_272k_tokens": 2.6e-08, + "cache_read_input_token_cost_above_272k_tokens_flex": 1.3e-08, }, "azure/gpt-6-astra": { "cache_read_input_token_cost": 1.25e-06, + "cache_read_input_token_cost_flex": 6.25e-07, "cache_creation_input_token_cost": 1.5625e-05, + "cache_creation_input_token_cost_flex": 7.8125e-06, "cache_creation_input_token_cost_above_272k_tokens": 3.125e-05, "input_cost_per_token": 1.25e-05, + "input_cost_per_token_flex": 6.25e-06, "litellm_provider": "azure", "max_input_tokens": 922000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 6.25e-05, + "output_cost_per_token_flex": 3.125e-05, "supports_function_calling": True, "supports_none_reasoning_effort": True, "input_cost_per_token_above_272k_tokens": 2.5e-05, @@ -1637,20 +1649,28 @@ COST_MAP_ENTRIES: Final[dict[str, dict[str, JsonValue]]] = { }, "azure/gpt-6.1-sol": { "cache_read_input_token_cost": 2.4e-07, + "cache_read_input_token_cost_flex": 6e-08, "cache_creation_input_token_cost": 3e-06, + "cache_creation_input_token_cost_flex": 1.5e-06, "cache_creation_input_token_cost_above_272k_tokens": 6e-06, + "cache_creation_input_token_cost_above_272k_tokens_flex": 3e-06, "input_cost_per_token": 2.4e-06, + "input_cost_per_token_flex": 1.2e-06, "litellm_provider": "azure", "max_input_tokens": 922000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 1.2e-05, + "output_cost_per_token_flex": 6e-06, "supports_function_calling": True, "supports_none_reasoning_effort": True, "input_cost_per_token_above_272k_tokens": 4.8e-06, + "input_cost_per_token_above_272k_tokens_flex": 2.4e-06, "output_cost_per_token_above_272k_tokens": 2.4e-05, + "output_cost_per_token_above_272k_tokens_flex": 9e-06, "cache_read_input_token_cost_above_272k_tokens": 4.8e-07, + "cache_read_input_token_cost_above_272k_tokens_flex": 1.2e-07, }, }