From 441f3bbe26b317285b38e8ba9487567385c55468 Mon Sep 17 00:00:00 2001 From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Fri, 21 Aug 2026 23:22:38 +0000 Subject: [PATCH] feat(databricks): add new Claude models and cached token pricing Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 222 +++++++++++++++++- model_prices_and_context_window.json | 222 +++++++++++++++++- .../databricks/test_databricks_pricing.py | 74 +++++- 3 files changed, 481 insertions(+), 37 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index a1961136d11..dbcf7819530 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -14502,6 +14502,8 @@ "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, "databricks/databricks-claude-3-7-sonnet": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14517,10 +14519,35 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, + "databricks/databricks-claude-fable-5": { + "cache_creation_input_token_cost": 1.2500075e-05, + "cache_read_input_token_cost": 1.0000060000000001e-06, + "input_cost_per_token": 1.000006e-05, + "input_dbu_cost_per_token": 0.000142858, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 5.0000020000000004e-05, + "output_dbu_cost_per_token": 0.000714286, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-haiku-4-5": { + "cache_creation_input_token_cost": 1.2500250000000002e-06, + "cache_read_input_token_cost": 1.0000200000000002e-07, "input_cost_per_token": 1.00002e-06, "input_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", @@ -14536,10 +14563,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4": { + "cache_creation_input_token_cost": 1.8750025e-05, + "cache_read_input_token_cost": 1.500002e-06, "input_cost_per_token": 1.5000020000000002e-05, "input_dbu_cost_per_token": 0.000214286, "litellm_provider": "databricks", @@ -14555,10 +14585,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4-1": { + "cache_creation_input_token_cost": 1.8750025e-05, + "cache_read_input_token_cost": 1.500002e-06, "input_cost_per_token": 1.5000020000000002e-05, "input_dbu_cost_per_token": 0.000214286, "litellm_provider": "databricks", @@ -14574,10 +14607,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4-5": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, "input_cost_per_token": 5.00003e-06, "input_dbu_cost_per_token": 7.1429e-05, "litellm_provider": "databricks", @@ -14593,11 +14629,14 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, "supports_output_config": true }, "databricks/databricks-claude-opus-4-6": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, "input_cost_per_token": 5.00003e-06, "input_dbu_cost_per_token": 7.1429e-05, "litellm_provider": "databricks", @@ -14613,10 +14652,82 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, + "databricks/databricks-claude-opus-4-7": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "databricks/databricks-claude-opus-4-8": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "databricks/databricks-claude-opus-5": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "databricks/databricks-claude-sonnet-4": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14632,10 +14743,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-1": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14651,10 +14765,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-5": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14670,10 +14787,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-6": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14689,10 +14809,36 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, + "databricks/databricks-claude-sonnet-5": { + "cache_creation_input_token_cost": 2.4999625e-06, + "cache_read_input_token_cost": 1.9999700000000004e-07, + "input_cost_per_token": 1.9999700000000004e-06, + "input_dbu_cost_per_token": 2.8571e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation. Claude Sonnet 5 DBU rates are Anthropic's introductory launch pricing, in effect through 2026-08-31, after which the standard Sonnet 5 rates (equal to Sonnet 4.5 / 4.6) take effect." + }, + "mode": "chat", + "output_cost_per_token": 9.999990000000002e-06, + "output_dbu_cost_per_token": 0.000142857, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "databricks/databricks-gemini-2-5-flash": { + "cache_creation_input_token_cost": 3.0001999999999996e-07, + "cache_read_input_token_cost": 3.0002e-08, "input_cost_per_token": 3.0001999999999996e-07, "input_dbu_cost_per_token": 4.285999999999999e-06, "litellm_provider": "databricks", @@ -14707,9 +14853,12 @@ "output_dbu_cost_per_token": 3.5714e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-2-5-pro": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14724,9 +14873,12 @@ "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-1-flash-lite": { + "cache_creation_input_token_cost": 3.1248e-07, + "cache_read_input_token_cost": 3.1248e-08, "input_cost_per_token": 3.1248e-07, "input_dbu_cost_per_token": 4.464e-06, "litellm_provider": "databricks", @@ -14741,9 +14893,12 @@ "output_dbu_cost_per_token": 2.6786e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-1-pro": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14758,9 +14913,12 @@ "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-flash": { + "cache_creation_input_token_cost": 6.250300000000001e-07, + "cache_read_input_token_cost": 6.250300000000001e-08, "input_cost_per_token": 6.2503e-07, "input_dbu_cost_per_token": 8.929e-06, "litellm_provider": "databricks", @@ -14775,9 +14933,12 @@ "output_dbu_cost_per_token": 5.3571e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-pro": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14792,6 +14953,7 @@ "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemma-3-12b": { @@ -14810,6 +14972,8 @@ "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, "databricks/databricks-gpt-5": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14822,9 +14986,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14837,9 +15004,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1-codex-max": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14852,9 +15022,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1-codex-mini": { + "cache_creation_input_token_cost": 2.4997000000000006e-07, + "cache_read_input_token_cost": 2.4997000000000005e-08, "input_cost_per_token": 2.4997e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", @@ -14867,9 +15040,12 @@ "mode": "chat", "output_cost_per_token": 1.99997e-06, "output_dbu_cost_per_token": 2.8571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-2": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14882,9 +15058,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-2-codex": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14897,9 +15076,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-3-codex": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14912,9 +15094,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14927,9 +15112,12 @@ "mode": "chat", "output_cost_per_token": 1.5000020000000002e-05, "output_dbu_cost_per_token": 0.000214286, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4-mini": { + "cache_creation_input_token_cost": 7.4998e-07, + "cache_read_input_token_cost": 7.499800000000001e-08, "input_cost_per_token": 7.4998e-07, "input_dbu_cost_per_token": 1.0714e-05, "litellm_provider": "databricks", @@ -14942,9 +15130,12 @@ "mode": "chat", "output_cost_per_token": 4.50002e-06, "output_dbu_cost_per_token": 6.4286e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4-nano": { + "cache_creation_input_token_cost": 1.9999e-07, + "cache_read_input_token_cost": 1.9999000000000003e-08, "input_cost_per_token": 1.9999e-07, "input_dbu_cost_per_token": 2.857e-06, "litellm_provider": "databricks", @@ -14957,9 +15148,12 @@ "mode": "chat", "output_cost_per_token": 1.24999e-06, "output_dbu_cost_per_token": 1.7857e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-mini": { + "cache_creation_input_token_cost": 2.4997000000000006e-07, + "cache_read_input_token_cost": 2.4997000000000005e-08, "input_cost_per_token": 2.4997000000000006e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", @@ -14972,9 +15166,12 @@ "mode": "chat", "output_cost_per_token": 1.9999700000000004e-06, "output_dbu_cost_per_token": 2.8571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-nano": { + "cache_creation_input_token_cost": 4.998e-08, + "cache_read_input_token_cost": 4.998000000000001e-09, "input_cost_per_token": 4.998e-08, "input_dbu_cost_per_token": 7.14e-07, "litellm_provider": "databricks", @@ -14987,7 +15184,8 @@ "mode": "chat", "output_cost_per_token": 3.9998000000000007e-07, "output_dbu_cost_per_token": 5.714000000000001e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-oss-120b": { "input_cost_per_token": 1.5000999999999998e-07, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index a1961136d11..dbcf7819530 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -14502,6 +14502,8 @@ "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, "databricks/databricks-claude-3-7-sonnet": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14517,10 +14519,35 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, + "databricks/databricks-claude-fable-5": { + "cache_creation_input_token_cost": 1.2500075e-05, + "cache_read_input_token_cost": 1.0000060000000001e-06, + "input_cost_per_token": 1.000006e-05, + "input_dbu_cost_per_token": 0.000142858, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 5.0000020000000004e-05, + "output_dbu_cost_per_token": 0.000714286, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-haiku-4-5": { + "cache_creation_input_token_cost": 1.2500250000000002e-06, + "cache_read_input_token_cost": 1.0000200000000002e-07, "input_cost_per_token": 1.00002e-06, "input_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", @@ -14536,10 +14563,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4": { + "cache_creation_input_token_cost": 1.8750025e-05, + "cache_read_input_token_cost": 1.500002e-06, "input_cost_per_token": 1.5000020000000002e-05, "input_dbu_cost_per_token": 0.000214286, "litellm_provider": "databricks", @@ -14555,10 +14585,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4-1": { + "cache_creation_input_token_cost": 1.8750025e-05, + "cache_read_input_token_cost": 1.500002e-06, "input_cost_per_token": 1.5000020000000002e-05, "input_dbu_cost_per_token": 0.000214286, "litellm_provider": "databricks", @@ -14574,10 +14607,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-opus-4-5": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, "input_cost_per_token": 5.00003e-06, "input_dbu_cost_per_token": 7.1429e-05, "litellm_provider": "databricks", @@ -14593,11 +14629,14 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, "supports_output_config": true }, "databricks/databricks-claude-opus-4-6": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, "input_cost_per_token": 5.00003e-06, "input_dbu_cost_per_token": 7.1429e-05, "litellm_provider": "databricks", @@ -14613,10 +14652,82 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, + "databricks/databricks-claude-opus-4-7": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "databricks/databricks-claude-opus-4-8": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "databricks/databricks-claude-opus-5": { + "cache_creation_input_token_cost": 6.2500375e-06, + "cache_read_input_token_cost": 5.000030000000001e-07, + "input_cost_per_token": 5.00003e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, + "mode": "chat", + "output_cost_per_token": 2.5000010000000002e-05, + "output_dbu_cost_per_token": 0.000357143, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "databricks/databricks-claude-sonnet-4": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14632,10 +14743,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-1": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14651,10 +14765,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-5": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14670,10 +14787,13 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, "databricks/databricks-claude-sonnet-4-6": { + "cache_creation_input_token_cost": 3.7499875e-06, + "cache_read_input_token_cost": 2.9999900000000006e-07, "input_cost_per_token": 2.9999900000000002e-06, "input_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", @@ -14689,10 +14809,36 @@ "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_assistant_prefill": true, "supports_function_calling": true, + "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true }, + "databricks/databricks-claude-sonnet-5": { + "cache_creation_input_token_cost": 2.4999625e-06, + "cache_read_input_token_cost": 1.9999700000000004e-07, + "input_cost_per_token": 1.9999700000000004e-06, + "input_dbu_cost_per_token": 2.8571e-05, + "litellm_provider": "databricks", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation. Claude Sonnet 5 DBU rates are Anthropic's introductory launch pricing, in effect through 2026-08-31, after which the standard Sonnet 5 rates (equal to Sonnet 4.5 / 4.6) take effect." + }, + "mode": "chat", + "output_cost_per_token": 9.999990000000002e-06, + "output_dbu_cost_per_token": 0.000142857, + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_assistant_prefill": false, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "databricks/databricks-gemini-2-5-flash": { + "cache_creation_input_token_cost": 3.0001999999999996e-07, + "cache_read_input_token_cost": 3.0002e-08, "input_cost_per_token": 3.0001999999999996e-07, "input_dbu_cost_per_token": 4.285999999999999e-06, "litellm_provider": "databricks", @@ -14707,9 +14853,12 @@ "output_dbu_cost_per_token": 3.5714e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-2-5-pro": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14724,9 +14873,12 @@ "output_dbu_cost_per_token": 0.000142857, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-1-flash-lite": { + "cache_creation_input_token_cost": 3.1248e-07, + "cache_read_input_token_cost": 3.1248e-08, "input_cost_per_token": 3.1248e-07, "input_dbu_cost_per_token": 4.464e-06, "litellm_provider": "databricks", @@ -14741,9 +14893,12 @@ "output_dbu_cost_per_token": 2.6786e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-1-pro": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14758,9 +14913,12 @@ "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-flash": { + "cache_creation_input_token_cost": 6.250300000000001e-07, + "cache_read_input_token_cost": 6.250300000000001e-08, "input_cost_per_token": 6.2503e-07, "input_dbu_cost_per_token": 8.929e-06, "litellm_provider": "databricks", @@ -14775,9 +14933,12 @@ "output_dbu_cost_per_token": 5.3571e-05, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemini-3-pro": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14792,6 +14953,7 @@ "output_dbu_cost_per_token": 0.000214286, "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", "supports_function_calling": true, + "supports_prompt_caching": true, "supports_tool_choice": true }, "databricks/databricks-gemma-3-12b": { @@ -14810,6 +14972,8 @@ "source": "https://www.databricks.com/product/pricing/foundation-model-serving" }, "databricks/databricks-gpt-5": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14822,9 +14986,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14837,9 +15004,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1-codex-max": { + "cache_creation_input_token_cost": 1.24999e-06, + "cache_read_input_token_cost": 1.24999e-07, "input_cost_per_token": 1.24999e-06, "input_dbu_cost_per_token": 1.7857e-05, "litellm_provider": "databricks", @@ -14852,9 +15022,12 @@ "mode": "chat", "output_cost_per_token": 9.999990000000002e-06, "output_dbu_cost_per_token": 0.000142857, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-1-codex-mini": { + "cache_creation_input_token_cost": 2.4997000000000006e-07, + "cache_read_input_token_cost": 2.4997000000000005e-08, "input_cost_per_token": 2.4997e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", @@ -14867,9 +15040,12 @@ "mode": "chat", "output_cost_per_token": 1.99997e-06, "output_dbu_cost_per_token": 2.8571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-2": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14882,9 +15058,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-2-codex": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14897,9 +15076,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-3-codex": { + "cache_creation_input_token_cost": 1.7500000000000002e-06, + "cache_read_input_token_cost": 1.7500000000000002e-07, "input_cost_per_token": 1.75e-06, "input_dbu_cost_per_token": 2.5e-05, "litellm_provider": "databricks", @@ -14912,9 +15094,12 @@ "mode": "chat", "output_cost_per_token": 1.4e-05, "output_dbu_cost_per_token": 0.0002, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4": { + "cache_creation_input_token_cost": 2.49998e-06, + "cache_read_input_token_cost": 2.49998e-07, "input_cost_per_token": 2.49998e-06, "input_dbu_cost_per_token": 3.5714e-05, "litellm_provider": "databricks", @@ -14927,9 +15112,12 @@ "mode": "chat", "output_cost_per_token": 1.5000020000000002e-05, "output_dbu_cost_per_token": 0.000214286, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4-mini": { + "cache_creation_input_token_cost": 7.4998e-07, + "cache_read_input_token_cost": 7.499800000000001e-08, "input_cost_per_token": 7.4998e-07, "input_dbu_cost_per_token": 1.0714e-05, "litellm_provider": "databricks", @@ -14942,9 +15130,12 @@ "mode": "chat", "output_cost_per_token": 4.50002e-06, "output_dbu_cost_per_token": 6.4286e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-4-nano": { + "cache_creation_input_token_cost": 1.9999e-07, + "cache_read_input_token_cost": 1.9999000000000003e-08, "input_cost_per_token": 1.9999e-07, "input_dbu_cost_per_token": 2.857e-06, "litellm_provider": "databricks", @@ -14957,9 +15148,12 @@ "mode": "chat", "output_cost_per_token": 1.24999e-06, "output_dbu_cost_per_token": 1.7857e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-mini": { + "cache_creation_input_token_cost": 2.4997000000000006e-07, + "cache_read_input_token_cost": 2.4997000000000005e-08, "input_cost_per_token": 2.4997000000000006e-07, "input_dbu_cost_per_token": 3.571e-06, "litellm_provider": "databricks", @@ -14972,9 +15166,12 @@ "mode": "chat", "output_cost_per_token": 1.9999700000000004e-06, "output_dbu_cost_per_token": 2.8571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-5-nano": { + "cache_creation_input_token_cost": 4.998e-08, + "cache_read_input_token_cost": 4.998000000000001e-09, "input_cost_per_token": 4.998e-08, "input_dbu_cost_per_token": 7.14e-07, "litellm_provider": "databricks", @@ -14987,7 +15184,8 @@ "mode": "chat", "output_cost_per_token": 3.9998000000000007e-07, "output_dbu_cost_per_token": 5.714000000000001e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving" + "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "supports_prompt_caching": true }, "databricks/databricks-gpt-oss-120b": { "input_cost_per_token": 1.5000999999999998e-07, diff --git a/tests/test_litellm/llms/databricks/test_databricks_pricing.py b/tests/test_litellm/llms/databricks/test_databricks_pricing.py index 1f8816f5076..df58e1c44c6 100644 --- a/tests/test_litellm/llms/databricks/test_databricks_pricing.py +++ b/tests/test_litellm/llms/databricks/test_databricks_pricing.py @@ -1,6 +1,29 @@ import json import os -import sys +from types import MappingProxyType +from typing import Final + +# (cache write, cache read) multipliers on the input rate, per model vendor, from +# https://www.databricks.com/product/pricing/proprietary-foundation-model-serving +CACHE_RATIOS: Final = MappingProxyType({"claude": (1.25, 0.1), "gemini": (1.0, 0.1), "gpt-5": (1.0, 0.1)}) + + +def _load_cost_map() -> dict: + json_path: Final = os.path.join(os.path.dirname(__file__), "../../../../model_prices_and_context_window.json") + assert os.path.exists(json_path), f"Could not find model_prices_and_context_window.json at {json_path}" + with open(json_path, "r") as f: + return json.load(f) + + +def _cache_vendor(model: str) -> str | None: + name: Final = model.removeprefix("databricks/databricks-") + if name.startswith("claude"): + return "claude" + if name.startswith("gemini"): + return "gemini" + if name.startswith("gpt-5") and "oss" not in name: + return "gpt-5" + return None def test_databricks_pricing_integrity(): @@ -8,14 +31,10 @@ def test_databricks_pricing_integrity(): Verifies that for all Databricks models in model_prices_and_context_window.json: USD Price == DBU Price * 0.07 """ - json_path = os.path.join( - os.path.dirname(__file__), "../../../../model_prices_and_context_window.json" - ) + json_path = os.path.join(os.path.dirname(__file__), "../../../../model_prices_and_context_window.json") # Verify file exists - assert os.path.exists( - json_path - ), f"Could not find model_prices_and_context_window.json at {json_path}" + assert os.path.exists(json_path), f"Could not find model_prices_and_context_window.json at {json_path}" with open(json_path, "r") as f: data = json.load(f) @@ -33,9 +52,7 @@ def test_databricks_pricing_integrity(): expected = input_dbu * conversion_rate # Allow small floating point difference if abs(input_usd - expected) > 1e-9: - errors.append( - f"{model} input mismatch: USD={input_usd}, DBU={input_dbu}, Expected={expected}" - ) + errors.append(f"{model} input mismatch: USD={input_usd}, DBU={input_dbu}, Expected={expected}") # Check Output Cost output_usd = info.get("output_cost_per_token") @@ -44,8 +61,39 @@ def test_databricks_pricing_integrity(): if output_usd is not None and output_dbu is not None: expected = output_dbu * conversion_rate if abs(output_usd - expected) > 1e-9: - errors.append( - f"{model} output mismatch: USD={output_usd}, DBU={output_dbu}, Expected={expected}" - ) + errors.append(f"{model} output mismatch: USD={output_usd}, DBU={output_dbu}, Expected={expected}") assert not errors, "\n" + "\n".join(errors) + + +def test_databricks_proprietary_models_have_cache_pricing(): + """ + Every Databricks-hosted Claude / Gemini / GPT-5 chat model must carry cached token pricing, + derived from its input rate with the vendor's cache write and cache read multipliers + """ + data: Final = _load_cost_map() + errors: Final[list[str]] = [] # mutable-ok: pytest failure message accumulator + + priced: Final[list[str]] = [] # mutable-ok: pytest failure message accumulator + for model, info in data.items(): + vendor: Final = _cache_vendor(model) if model.startswith("databricks/") else None + if vendor is None or info.get("mode") != "chat": + continue + priced.append(model) + write_ratio, read_ratio = CACHE_RATIOS[vendor] + input_usd = info["input_cost_per_token"] + for field, ratio in ( + ("cache_creation_input_token_cost", write_ratio), + ("cache_read_input_token_cost", read_ratio), + ): + actual = info.get(field) + expected = input_usd * ratio + if actual is None: + errors.append(f"{model} missing {field}") + elif abs(actual - expected) > 1e-12: + errors.append(f"{model} {field} mismatch: got {actual}, expected {expected}") + if not info.get("supports_prompt_caching"): + errors.append(f"{model} missing supports_prompt_caching") + + assert priced, "no Databricks proprietary chat models found" + assert not errors, "\n" + "\n".join(errors)