diff --git a/tests/integration/cost_calculation/cost_tracking_cases.json b/tests/integration/cost_calculation/cost_tracking_cases.json index 81beff68c7c..9e22212249c 100644 --- a/tests/integration/cost_calculation/cost_tracking_cases.json +++ b/tests/integration/cost_calculation/cost_tracking_cases.json @@ -642,20 +642,20 @@ "gpt-realtime-mini-2025-12-15": { "litellm_provider": "openai", "mode": "realtime", - "input_cost_per_token": 6.0e-07, + "input_cost_per_token": 6e-07, "output_cost_per_token": 2.4e-06, - "input_cost_per_audio_token": 1.0e-05, - "cache_read_input_token_cost": 6.0e-08, - "cache_read_input_audio_token_cost": 3.0e-07, - "output_cost_per_audio_token": 2.0e-05 + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 6e-08, + "cache_read_input_audio_token_cost": 3e-07, + "output_cost_per_audio_token": 2e-05 }, "gpt-realtime-2.1": { "litellm_provider": "openai", "mode": "realtime", - "input_cost_per_token": 4.0e-06, + "input_cost_per_token": 4e-06, "input_cost_per_audio_token": 3.2e-05, - "cache_read_input_token_cost": 4.0e-07, - "cache_read_input_audio_token_cost": 4.0e-07, + "cache_read_input_token_cost": 4e-07, + "cache_read_input_audio_token_cost": 4e-07, "output_cost_per_token": 2.4e-05, "output_cost_per_audio_token": 6.4e-05 }, @@ -769,6 +769,52 @@ "search_context_size_medium": 0.014 }, "web_search_billing_unit": "per_query" + }, + "xai/grok-4.7": { + "litellm_provider": "xai", + "mode": "chat", + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, + "cache_read_input_token_cost": 5e-07, + "max_tokens": 500000, + "max_input_tokens": 500000, + "max_output_tokens": 500000, + "supports_function_calling": true + }, + "us.anthropic.claude-opus-5-5": { + "litellm_provider": "bedrock_converse", + "mode": "chat", + "input_cost_per_token": 4.4e-06, + "output_cost_per_token": 2.2e-05, + "cache_read_input_token_cost": 2.2e-07, + "cache_creation_input_token_cost": 5.5e-06, + "max_tokens": 128000, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "supports_function_calling": true + }, + "claude-opus-5-5": { + "litellm_provider": "anthropic", + "mode": "chat", + "input_cost_per_token": 4e-06, + "output_cost_per_token": 2e-05, + "cache_read_input_token_cost": 2e-07, + "cache_creation_input_token_cost": 5e-06, + "max_tokens": 128000, + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "supports_function_calling": true + }, + "gpt-6.1-sol": { + "litellm_provider": "openai", + "mode": "chat", + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1e-07, + "max_tokens": 128000, + "max_input_tokens": 922000, + "max_output_tokens": 128000, + "supports_function_calling": true } }, "cases": [ @@ -29687,7 +29733,13 @@ } } }, - "expected": {"spend": 0.0025118, "input_cost": 0.0020792, "output_cost": 0.0004326, "prompt_tokens": 1840, "completion_tokens": 412} + "expected": { + "spend": 0.0025118, + "input_cost": 0.0020792, + "output_cost": 0.0004326, + "prompt_tokens": 1840, + "completion_tokens": 412 + } }, { "name": "deepseek-deepseek-v4-chat-prompt_cache_hit", @@ -32102,6 +32154,190 @@ "prompt_tokens": 100, "completion_tokens": 100 } + }, + { + "name": "xai-grok-4.7-input_text", + "covers": "quota_management.spend_tracking.cost_matrix.logs_cost", + "model": "xai/grok-4.7", + "request": { + "model": "$MODEL", + "messages": [ + { + "role": "user", + "content": "Say hello." + } + ], + "stream": false, + "allowed_openai_params": [] + }, + "response": { + "content_type": "application/json", + "body": { + "id": "chatcmpl-$REQUEST_ID", + "object": "chat.completion", + "model": "grok-4.7", + "choices": [ + { + "index": 0, + "message": { + "role": "assistant", + "content": "Hello." + }, + "finish_reason": "stop" + } + ], + "usage": { + "prompt_tokens": 1840, + "completion_tokens": 412, + "total_tokens": 2252 + } + } + }, + "expected": { + "spend": 0.006152, + "input_cost": 0.00368, + "output_cost": 0.002472, + "prompt_tokens": 1840, + "completion_tokens": 412 + } + }, + { + "name": "us.anthropic.claude-opus-5-5-input_text", + "covers": "quota_management.spend_tracking.cost_matrix.logs_cost", + "model": "us.anthropic.claude-opus-5-5", + "request": { + "model": "$MODEL", + "messages": [ + { + "role": "user", + "content": "Say hello." + } + ], + "stream": false, + "allowed_openai_params": [] + }, + "response": { + "content_type": "application/json", + "body": { + "output": { + "message": { + "role": "assistant", + "content": [ + { + "text": "Hello." + } + ] + } + }, + "stopReason": "end_turn", + "usage": { + "inputTokens": 1840, + "outputTokens": 412, + "totalTokens": 2252 + }, + "metrics": { + "latencyMs": 42 + } + } + }, + "expected": { + "spend": 0.01716, + "input_cost": 0.008096, + "output_cost": 0.009064, + "prompt_tokens": 1840, + "completion_tokens": 412 + } + }, + { + "name": "claude-opus-5-5-input_text", + "covers": "quota_management.spend_tracking.cost_matrix.logs_cost", + "model": "claude-opus-5-5", + "request": { + "model": "$MODEL", + "messages": [ + { + "role": "user", + "content": "Say hello." + } + ], + "stream": false, + "allowed_openai_params": [] + }, + "response": { + "content_type": "application/json", + "body": { + "id": "msg_$REQUEST_ID", + "type": "message", + "role": "assistant", + "model": "claude-opus-5-5", + "content": [ + { + "type": "text", + "text": "Hello." + } + ], + "stop_reason": "end_turn", + "stop_sequence": null, + "usage": { + "input_tokens": 1840, + "output_tokens": 412 + } + } + }, + "expected": { + "spend": 0.0156, + "input_cost": 0.00736, + "output_cost": 0.00824, + "prompt_tokens": 1840, + "completion_tokens": 412 + } + }, + { + "name": "gpt-6.1-sol-input_text", + "covers": "quota_management.spend_tracking.cost_matrix.logs_cost", + "model": "gpt-6.1-sol", + "request": { + "model": "$MODEL", + "messages": [ + { + "role": "user", + "content": "Say hello." + } + ], + "stream": false, + "allowed_openai_params": [] + }, + "response": { + "content_type": "application/json", + "body": { + "id": "chatcmpl-$REQUEST_ID", + "object": "chat.completion", + "created": 1, + "model": "gpt-6.1-sol", + "choices": [ + { + "index": 0, + "message": { + "role": "assistant", + "content": "Hello." + }, + "finish_reason": "stop" + } + ], + "usage": { + "prompt_tokens": 1840, + "completion_tokens": 412, + "total_tokens": 2252 + } + } + }, + "expected": { + "spend": 0.0078, + "input_cost": 0.00368, + "output_cost": 0.00412, + "prompt_tokens": 1840, + "completion_tokens": 412 + } } ], "batch_cases": [ @@ -32149,8 +32385,8 @@ ], "expected": { "spend": 0.000126, - "input_cost": 0.000056, - "output_cost": 0.00007, + "input_cost": 5.6e-05, + "output_cost": 7e-05, "prompt_tokens": 100, "completion_tokens": 10, "cost_header": false diff --git a/tests/integration/cost_calculation/stream_parity/__init__.py b/tests/integration/cost_calculation/stream_parity/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/integration/cost_calculation/stream_parity/bases/__init__.py b/tests/integration/cost_calculation/stream_parity/bases/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/integration/cost_calculation/stream_parity/bases/anthropic.py b/tests/integration/cost_calculation/stream_parity/bases/anthropic.py new file mode 100644 index 00000000000..347e6b17b4f --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/bases/anthropic.py @@ -0,0 +1,123 @@ +from typing import Final + +from integration.cost_calculation.cost_tracking_case import ( + CostTrackingTestCase, + Deployment, + ExactExpected, + JsonResponse, + SseResponse, +) +from integration.cost_calculation.stream_parity.case import COVERS, MODEL, REQUEST_ID, StreamParityTestCase, sse_frames + +CLAUDE_OPUS_5_5_MESSAGES: Final = CostTrackingTestCase( + name="claude-opus-5-5-messages-parity", + covers=COVERS, + model="claude-opus-5-5", + endpoint="/v1/messages", + deployment=Deployment(input_cost_per_token=0.001, output_cost_per_token=0.002), + request={ + "model": MODEL, + "max_tokens": 64, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + response=JsonResponse( + content_type="application/json", + body={ + "id": f"msg_{REQUEST_ID}", + "type": "message", + "role": "assistant", + "model": "claude-opus-5-5", + "content": [{"type": "text", "text": "Hello."}], + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 30, "output_tokens": 40}, + }, + ), + expected=ExactExpected(spend=0.11, input_cost=0.03, output_cost=0.08, prompt_tokens=30, completion_tokens=40), +) + +CLAUDE_OPUS_5_5_STREAM_RESPONSE: Final = SseResponse( + content_type="text/event-stream", + frames=sse_frames( + { + "type": "message_start", + "message": { + "id": f"msg_{REQUEST_ID}", + "type": "message", + "role": "assistant", + "model": "claude-opus-5-5", + "content": [], + "stop_reason": None, + "stop_sequence": None, + "usage": {"input_tokens": 30, "output_tokens": 1}, + }, + }, + {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}, + {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Hello."}}, + {"type": "content_block_stop", "index": 0}, + { + "type": "message_delta", + "delta": {"stop_reason": "end_turn", "stop_sequence": None}, + "usage": {"output_tokens": 40}, + }, + {"type": "message_stop"}, + ), +) + +CLAUDE_OPUS_5_5_MESSAGES_PARITY: Final = StreamParityTestCase( + plain=CLAUDE_OPUS_5_5_MESSAGES, + streamed=CLAUDE_OPUS_5_5_MESSAGES.model_copy( + update={ + "name": "claude-opus-5-5-messages-parity-stream", + "request": {**CLAUDE_OPUS_5_5_MESSAGES.request, "stream": True}, + "response": CLAUDE_OPUS_5_5_STREAM_RESPONSE, + } + ), +) + +CLAUDE_OPUS_5_5_CHAT_COMPLETIONS: Final = CLAUDE_OPUS_5_5_MESSAGES.model_copy( + update={ + "name": "claude-opus-5-5-chat-parity", + "endpoint": "/v1/chat/completions", + "request": { + "model": MODEL, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + } +) + +CLAUDE_OPUS_5_5_CHAT_COMPLETIONS_PARITY: Final = StreamParityTestCase( + plain=CLAUDE_OPUS_5_5_CHAT_COMPLETIONS, + streamed=CLAUDE_OPUS_5_5_CHAT_COMPLETIONS.model_copy( + update={ + "name": "claude-opus-5-5-chat-parity-stream", + "request": { + **CLAUDE_OPUS_5_5_CHAT_COMPLETIONS.request, + "stream": True, + "stream_options": {"include_usage": True}, + }, + "response": CLAUDE_OPUS_5_5_STREAM_RESPONSE, + } + ), +) + +CLAUDE_OPUS_5_5_RESPONSES: Final = CLAUDE_OPUS_5_5_MESSAGES.model_copy( + update={ + "name": "claude-opus-5-5-responses-parity", + "endpoint": "/v1/responses", + "request": {"model": MODEL, "input": "Say hello.", "cache": {"no-cache": True}}, + } +) + +CLAUDE_OPUS_5_5_RESPONSES_PARITY: Final = StreamParityTestCase( + plain=CLAUDE_OPUS_5_5_RESPONSES, + streamed=CLAUDE_OPUS_5_5_RESPONSES.model_copy( + update={ + "name": "claude-opus-5-5-responses-parity-stream", + "request": {**CLAUDE_OPUS_5_5_RESPONSES.request, "stream": True}, + "response": CLAUDE_OPUS_5_5_STREAM_RESPONSE, + } + ), +) diff --git a/tests/integration/cost_calculation/stream_parity/bases/bedrock.py b/tests/integration/cost_calculation/stream_parity/bases/bedrock.py new file mode 100644 index 00000000000..501cb5e8cab --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/bases/bedrock.py @@ -0,0 +1,62 @@ +"""/v1/messages on a Bedrock Converse deployment: the messages -> converse bridge.""" + +from typing import Final + +from integration.cost_calculation.cost_tracking_case import ( + CostTrackingTestCase, + Deployment, + EventStreamEvent, + EventStreamResponse, + ExactExpected, + JsonResponse, +) +from integration.cost_calculation.stream_parity.case import COVERS, MODEL, StreamParityTestCase + +_USAGE: Final = {"inputTokens": 30, "outputTokens": 40, "totalTokens": 70} + +CLAUDE_OPUS_5_5_CONVERSE_MESSAGES: Final = CostTrackingTestCase( + name="bedrock-converse-claude-opus-5-5-messages-parity", + covers=COVERS, + model="us.anthropic.claude-opus-5-5", + endpoint="/v1/messages", + deployment=Deployment(input_cost_per_token=0.001, output_cost_per_token=0.002), + request={ + "model": MODEL, + "max_tokens": 64, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + response=JsonResponse( + content_type="application/json", + body={ + "output": {"message": {"role": "assistant", "content": [{"text": "Hello."}]}}, + "stopReason": "end_turn", + "usage": _USAGE, + "metrics": {"latencyMs": 1}, + }, + ), + expected=ExactExpected(spend=0.11, input_cost=0.03, output_cost=0.08, prompt_tokens=30, completion_tokens=40), +) + +CLAUDE_OPUS_5_5_CONVERSE_MESSAGES_PARITY: Final = StreamParityTestCase( + plain=CLAUDE_OPUS_5_5_CONVERSE_MESSAGES, + streamed=CLAUDE_OPUS_5_5_CONVERSE_MESSAGES.model_copy( + update={ + "name": "bedrock-converse-claude-opus-5-5-messages-parity-stream", + "request": {**CLAUDE_OPUS_5_5_CONVERSE_MESSAGES.request, "stream": True}, + "response": EventStreamResponse( + content_type="application/vnd.amazon.eventstream", + framing="converse", + events=( + EventStreamEvent(event_type="messageStart", payload={"role": "assistant"}), + EventStreamEvent( + event_type="contentBlockDelta", payload={"delta": {"text": "Hello."}, "contentBlockIndex": 0} + ), + EventStreamEvent(event_type="contentBlockStop", payload={"contentBlockIndex": 0}), + EventStreamEvent(event_type="messageStop", payload={"stopReason": "end_turn"}), + EventStreamEvent(event_type="metadata", payload={"usage": _USAGE, "metrics": {"latencyMs": 1}}), + ), + ), + } + ), +) diff --git a/tests/integration/cost_calculation/stream_parity/bases/openai.py b/tests/integration/cost_calculation/stream_parity/bases/openai.py new file mode 100644 index 00000000000..ef94094e9f0 --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/bases/openai.py @@ -0,0 +1,91 @@ +from typing import Final + +from integration.cost_calculation.cost_tracking_case import ( + CostTrackingTestCase, + Deployment, + ExactExpected, + JsonResponse, + SseResponse, +) +from integration.cost_calculation.stream_parity.bases.openai_responses import ( + GPT_6_1_SOL_RESPONSES, + GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE, +) +from integration.cost_calculation.stream_parity.case import COVERS, MODEL, REQUEST_ID, StreamParityTestCase, sse_frames + +_USAGE: Final = {"prompt_tokens": 30, "completion_tokens": 40, "total_tokens": 70} +_CHUNK: Final = { + "id": f"chatcmpl-{REQUEST_ID}", + "object": "chat.completion.chunk", + "created": 1, + "model": "gpt-6.1-sol", +} + +GPT_6_1_SOL_CHAT_COMPLETIONS: Final = CostTrackingTestCase( + name="gpt-6.1-sol-chat-parity", + covers=COVERS, + model="gpt-6.1-sol", + endpoint="/v1/chat/completions", + deployment=Deployment(input_cost_per_token=0.001, output_cost_per_token=0.002), + request={"model": MODEL, "messages": [{"role": "user", "content": "Say hello."}], "cache": {"no-cache": True}}, + response=JsonResponse( + content_type="application/json", + body={ + "id": f"chatcmpl-{REQUEST_ID}", + "object": "chat.completion", + "created": 1, + "model": "gpt-6.1-sol", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "Hello."}, "finish_reason": "stop"}], + "usage": _USAGE, + }, + ), + expected=ExactExpected(spend=0.11, input_cost=0.03, output_cost=0.08, prompt_tokens=30, completion_tokens=40), +) + +GPT_6_1_SOL_CHAT_COMPLETIONS_PARITY: Final = StreamParityTestCase( + plain=GPT_6_1_SOL_CHAT_COMPLETIONS, + streamed=GPT_6_1_SOL_CHAT_COMPLETIONS.model_copy( + update={ + "name": "gpt-6.1-sol-chat-parity-stream", + "request": { + **GPT_6_1_SOL_CHAT_COMPLETIONS.request, + "stream": True, + "stream_options": {"include_usage": True}, + }, + "response": SseResponse( + content_type="text/event-stream", + frames=sse_frames( + {**_CHUNK, "choices": [{"index": 0, "delta": {"role": "assistant", "content": "Hello."}}]}, + {**_CHUNK, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}, + {**_CHUNK, "choices": [], "usage": _USAGE}, + done=True, + ), + ), + } + ), +) + +GPT_6_1_SOL_MESSAGES: Final = GPT_6_1_SOL_RESPONSES.model_copy( + update={ + "name": "gpt-6.1-sol-messages-parity", + "model": "gpt-6.1-sol", + "endpoint": "/v1/messages", + "request": { + "model": MODEL, + "max_tokens": 64, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + } +) + +GPT_6_1_SOL_MESSAGES_PARITY: Final = StreamParityTestCase( + plain=GPT_6_1_SOL_MESSAGES, + streamed=GPT_6_1_SOL_MESSAGES.model_copy( + update={ + "name": "gpt-6.1-sol-messages-parity-stream", + "request": {**GPT_6_1_SOL_MESSAGES.request, "stream": True}, + "response": GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE, + } + ), +) diff --git a/tests/integration/cost_calculation/stream_parity/bases/openai_responses.py b/tests/integration/cost_calculation/stream_parity/bases/openai_responses.py new file mode 100644 index 00000000000..93a47ed74bf --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/bases/openai_responses.py @@ -0,0 +1,115 @@ +from typing import Final + +from integration.cost_calculation.cost_tracking_case import ( + CostTrackingTestCase, + Deployment, + ExactExpected, + JsonResponse, + SseResponse, +) +from integration.cost_calculation.stream_parity.case import COVERS, MODEL, REQUEST_ID, StreamParityTestCase, sse_frames + +_ITEM: Final = { + "id": "msg_1", + "type": "message", + "role": "assistant", + "status": "completed", + "content": [{"type": "output_text", "text": "Hello.", "annotations": []}], +} +_RESPONSE: Final = { + "id": f"resp_{REQUEST_ID}", + "object": "response", + "created_at": 1, + "status": "completed", + "model": "gpt-6.1-sol", + "output": [_ITEM], + "usage": {"input_tokens": 30, "output_tokens": 40, "total_tokens": 70}, +} + +GPT_6_1_SOL_RESPONSES: Final = CostTrackingTestCase( + name="gpt-6.1-sol-responses-parity", + covers=COVERS, + model="gpt-6.1-sol", + endpoint="/v1/responses", + deployment=Deployment(input_cost_per_token=0.001, output_cost_per_token=0.002), + request={"model": MODEL, "input": "Say hello.", "cache": {"no-cache": True}}, + response=JsonResponse(content_type="application/json", body=_RESPONSE), + expected=ExactExpected(spend=0.11, input_cost=0.03, output_cost=0.08, prompt_tokens=30, completion_tokens=40), +) + +GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE: Final = SseResponse( + content_type="text/event-stream", + frames=sse_frames( + { + "type": "response.created", + "sequence_number": 0, + "response": {**_RESPONSE, "status": "in_progress", "output": []}, + }, + { + "type": "response.output_item.added", + "sequence_number": 1, + "output_index": 0, + "item": {**_ITEM, "status": "in_progress", "content": []}, + }, + { + "type": "response.output_text.delta", + "sequence_number": 2, + "item_id": "msg_1", + "output_index": 0, + "content_index": 0, + "delta": "Hello.", + }, + {"type": "response.output_item.done", "sequence_number": 3, "output_index": 0, "item": _ITEM}, + {"type": "response.completed", "sequence_number": 4, "response": _RESPONSE}, + ), +) + +GPT_6_1_SOL_RESPONSES_PARITY: Final = StreamParityTestCase( + plain=GPT_6_1_SOL_RESPONSES, + streamed=GPT_6_1_SOL_RESPONSES.model_copy( + update={ + "name": "gpt-6.1-sol-responses-parity-stream", + "request": {**GPT_6_1_SOL_RESPONSES.request, "stream": True}, + "response": GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE, + } + ), +) + +_PRO_RESPONSE: Final = {**_RESPONSE, "model": "gpt-5.5-pro"} +GPT_5_5_PRO_STREAM_RESPONSE: Final = GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE.model_copy( + update={ + "frames": tuple( + frame.replace('"model": "gpt-6.1-sol"', '"model": "gpt-5.5-pro"') + for frame in GPT_6_1_SOL_RESPONSES_PARITY_STREAM_RESPONSE.frames + ) + } +) + +GPT_5_5_PRO_CHAT_COMPLETIONS: Final = GPT_6_1_SOL_RESPONSES.model_copy( + update={ + "name": "gpt-5.5-pro-chat-parity", + "model": "gpt-5.5-pro", + "response": JsonResponse(content_type="application/json", body=_PRO_RESPONSE), + "endpoint": "/v1/chat/completions", + "request": { + "model": MODEL, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + } +) + +GPT_5_5_PRO_CHAT_COMPLETIONS_PARITY: Final = StreamParityTestCase( + plain=GPT_5_5_PRO_CHAT_COMPLETIONS, + streamed=GPT_5_5_PRO_CHAT_COMPLETIONS.model_copy( + update={ + "name": "gpt-5.5-pro-chat-parity-stream", + "request": { + **GPT_5_5_PRO_CHAT_COMPLETIONS.request, + "stream": True, + "stream_options": {"include_usage": True}, + }, + "response": GPT_5_5_PRO_STREAM_RESPONSE, + } + ), +) diff --git a/tests/integration/cost_calculation/stream_parity/bases/xai.py b/tests/integration/cost_calculation/stream_parity/bases/xai.py new file mode 100644 index 00000000000..2ab0a5a5c4a --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/bases/xai.py @@ -0,0 +1,65 @@ +"""/v1/messages on a chat-completions provider: the messages -> chat/completions bridge.""" + +from typing import Final + +from integration.cost_calculation.cost_tracking_case import ( + CostTrackingTestCase, + Deployment, + ExactExpected, + JsonResponse, + SseResponse, +) +from integration.cost_calculation.stream_parity.case import COVERS, MODEL, REQUEST_ID, StreamParityTestCase, sse_frames + +_USAGE: Final = {"prompt_tokens": 30, "completion_tokens": 40, "total_tokens": 70} +_CHUNK: Final = { + "id": f"chatcmpl-{REQUEST_ID}", + "object": "chat.completion.chunk", + "created": 1, + "model": "grok-4.7", +} + +GROK_4_7_MESSAGES: Final = CostTrackingTestCase( + name="grok-4.7-messages-parity", + covers=COVERS, + model="xai/grok-4.7", + endpoint="/v1/messages", + deployment=Deployment(input_cost_per_token=0.001, output_cost_per_token=0.002), + request={ + "model": MODEL, + "max_tokens": 64, + "messages": [{"role": "user", "content": "Say hello."}], + "cache": {"no-cache": True}, + }, + response=JsonResponse( + content_type="application/json", + body={ + "id": f"chatcmpl-{REQUEST_ID}", + "object": "chat.completion", + "created": 1, + "model": "grok-4.7", + "choices": [{"index": 0, "message": {"role": "assistant", "content": "Hello."}, "finish_reason": "stop"}], + "usage": _USAGE, + }, + ), + expected=ExactExpected(spend=0.11, input_cost=0.03, output_cost=0.08, prompt_tokens=30, completion_tokens=40), +) + +GROK_4_7_MESSAGES_PARITY: Final = StreamParityTestCase( + plain=GROK_4_7_MESSAGES, + streamed=GROK_4_7_MESSAGES.model_copy( + update={ + "name": "grok-4.7-messages-parity-stream", + "request": {**GROK_4_7_MESSAGES.request, "stream": True}, + "response": SseResponse( + content_type="text/event-stream", + frames=sse_frames( + {**_CHUNK, "choices": [{"index": 0, "delta": {"role": "assistant", "content": "Hello."}}]}, + {**_CHUNK, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}, + {**_CHUNK, "choices": [], "usage": _USAGE}, + done=True, + ), + ), + } + ), +) diff --git a/tests/integration/cost_calculation/stream_parity/case.py b/tests/integration/cost_calculation/stream_parity/case.py new file mode 100644 index 00000000000..1eb7c598939 --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/case.py @@ -0,0 +1,44 @@ +import json +from collections.abc import Mapping +from dataclasses import dataclass +from typing import Final + +from integration.cost_calculation.cost_tracking_case import CostTrackingTestCase +from pydantic import JsonValue + +COVERS: Final = "quota_management.spend_tracking.scripted_wire.logs_cost" +REQUEST_ID: Final = "$REQUEST_ID" +MODEL: Final = "$MODEL" + + +def sse_frames(*events: Mapping[str, JsonValue], done: bool = False) -> tuple[str, ...]: + frames: Final = tuple( + f"event: {event['type']}\ndata: {json.dumps(event)}" if "type" in event else f"data: {json.dumps(event)}" + for event in events + ) + return (*frames, "data: [DONE]") if done else frames + + +@dataclass(frozen=True, slots=True, kw_only=True) +class StreamParityTestCase: + """The same request as two CostTrackingTestCases, `plain` answered with JSON and `streamed` answered with SSE, + that must both bill the one `expected` row.""" + + plain: CostTrackingTestCase + streamed: CostTrackingTestCase + + def __post_init__(self) -> None: + if self.plain.expected != self.streamed.expected: + raise ValueError(f"{self.id}: plain and streamed cases expect different rows") + if self.streamed.request.get("stream") is not True or self.plain.request.get("stream") is True: + raise ValueError(f"{self.id}: streamed case must set stream: true and the plain case must not") + if (self.plain.model, self.plain.endpoint, self.plain.deployment) != ( + self.streamed.model, + self.streamed.endpoint, + self.streamed.deployment, + ): + raise ValueError(f"{self.id}: plain and streamed cases must share model, endpoint and deployment") + + @property + def id(self) -> str: + return self.plain.name diff --git a/tests/integration/cost_calculation/stream_parity/runner.py b/tests/integration/cost_calculation/stream_parity/runner.py new file mode 100644 index 00000000000..17ce7058c14 --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/runner.py @@ -0,0 +1,70 @@ +from collections import Counter +from hashlib import sha256 +from typing import Final + +from integration._support.client import Gateway, Scenario +from integration.cost_calculation.assertions import assert_exact +from integration.cost_calculation.conftest import ( + poll_rows, + poll_rows_where, + read_rows_now, + register_scenario_deployment, +) +from integration.cost_calculation.cost_tracking_case import ExactExpected +from integration.cost_calculation.stream_parity.bases import openai_responses +from integration.cost_calculation.stream_parity.case import StreamParityTestCase + +FLUSH_CASE: Final = openai_responses.GPT_6_1_SOL_RESPONSES + + +def flush_spend_logs(scenario: Scenario, gateway: Gateway, key: str) -> str: + """A /v1/responses row wakes the spend-log writer at once, and the queue is FIFO, so every row queued before it lands.""" + flush: Final = register_scenario_deployment(scenario, FLUSH_CASE, "flush", key) + response: Final = gateway.request( + "POST", FLUSH_CASE.endpoint, {**FLUSH_CASE.request, "model": flush.model_name}, key=key + ) + assert response.status_code == 200, ("flush", response.text[:400]) + poll_rows_where(key, 1, lambda row: row.model_id == flush.identity) + return flush.identity + + +def assert_stream_parity(case: StreamParityTestCase, gateway: Gateway) -> None: + expected: Final = case.plain.expected + assert isinstance(expected, ExactExpected), case.id + with gateway.scenario() as scenario: + key: Final = scenario.key() + deployments: Final = { + leg.name: register_scenario_deployment(scenario, leg, sha256(leg.name.encode()).hexdigest()[:12], key) + for leg in (case.plain, case.streamed) + } + responses: Final = { + leg.name: gateway.request( + "POST", leg.endpoint, {**leg.request, "model": deployments[leg.name].model_name}, key=key + ) + for leg in (case.plain, case.streamed) + } + for leg in (case.plain, case.streamed): + assert responses[leg.name].status_code == 200, (leg.name, responses[leg.name].text[:400]) + + poll_rows(key, 2) + flush_identity: Final = flush_spend_logs(scenario, gateway, key) + rows: Final = read_rows_now(key) + rows_per_deployment: Final = Counter(row.model_id for row in rows) + expected_rows: Final = {deployments[leg.name].identity: 1 for leg in (case.plain, case.streamed)} + assert rows_per_deployment == {**expected_rows, flush_identity: 1}, ( + f"{case.id}: expected exactly one SpendLogs row per leg plus the flush row, got {dict(rows_per_deployment)}", + rows, + ) + billed: Final = { + leg.name: next(row for row in rows if row.model_id == deployments[leg.name].identity) + for leg in (case.plain, case.streamed) + } + for leg in (case.plain, case.streamed): + assert_exact(leg.name, leg.response.content_type, expected, billed[leg.name], responses[leg.name]) + plain: Final = billed[case.plain.name] + streamed: Final = billed[case.streamed.name] + assert (streamed.spend, streamed.prompt_tokens, streamed.completion_tokens) == ( + plain.spend, + plain.prompt_tokens, + plain.completion_tokens, + ), (plain, streamed) diff --git a/tests/integration/cost_calculation/stream_parity/test_basic.py b/tests/integration/cost_calculation/stream_parity/test_basic.py new file mode 100644 index 00000000000..00089f9bb26 --- /dev/null +++ b/tests/integration/cost_calculation/stream_parity/test_basic.py @@ -0,0 +1,40 @@ +"""A streamed request bills exactly what the same request bills without streaming. + +Streaming reconstructs the response, its usage and its model name from chunks after the fact, in a separate code +path from the non-streamed one (LIT-6241, LIT-6872, LIT-7632, LIT-7729, LIT-9044, LIT-9065). Each case is a plain and +a streamed CostTrackingTestCase for one request, sent through the same key, and both LiteLLM_SpendLogs rows must +bill the case's expected row. + +Three cells are the native routes (chat, /v1/messages, /v1/responses); the rest are the bridges that translate one +API into another before the provider call, each with its own stream reassembly +""" + +from typing import Final + +import pytest +from integration._support.client import Gateway +from integration.cost_calculation.stream_parity.bases import anthropic, bedrock, openai, openai_responses, xai +from integration.cost_calculation.stream_parity.case import StreamParityTestCase +from integration.cost_calculation.stream_parity.runner import assert_stream_parity + +NATIVE: Final = ( + openai.GPT_6_1_SOL_CHAT_COMPLETIONS_PARITY, + anthropic.CLAUDE_OPUS_5_5_MESSAGES_PARITY, + openai_responses.GPT_6_1_SOL_RESPONSES_PARITY, +) +BRIDGES: Final = ( + xai.GROK_4_7_MESSAGES_PARITY, + openai.GPT_6_1_SOL_MESSAGES_PARITY, + bedrock.CLAUDE_OPUS_5_5_CONVERSE_MESSAGES_PARITY, + openai_responses.GPT_5_5_PRO_CHAT_COMPLETIONS_PARITY, + anthropic.CLAUDE_OPUS_5_5_RESPONSES_PARITY, +) +CASES: Final = (*NATIVE, *BRIDGES) + + +@pytest.mark.parametrize("case", CASES, ids=[case.id for case in CASES]) +@pytest.mark.timeout(180) +def test_streamed_request_bills_the_same_row_as_the_non_streamed_request( + case: StreamParityTestCase, gateway: Gateway +) -> None: + assert_stream_parity(case, gateway)