From f0849eb0c94948cb013274ab6875c7b0a5988c0f Mon Sep 17 00:00:00 2001 From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Date: Sat, 29 Aug 2026 13:11:26 +0000 Subject: [PATCH] fix(models): xai retirement repricing, bedrock grok-4.6 caching, openai/gemini deprecation dates Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> --- ...odel_prices_and_context_window_backup.json | 188 +++++++++++------- model_prices_and_context_window.json | 188 +++++++++++------- .../test_anthropic_cache_control_hook.py | 12 ++ .../llms/xai/test_xai_cost_calculator.py | 41 ++-- .../xai/test_xai_redirected_slug_pricing.py | 133 +++++++++++++ 5 files changed, 388 insertions(+), 174 deletions(-) create mode 100644 tests/test_litellm/llms/xai/test_xai_redirected_slug_pricing.py diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index ab1054d985b..390489a76f6 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -23096,7 +23096,8 @@ "supports_system_messages": true, "supports_video_input": true, "supports_vision": true, - "tpm": 800000 + "tpm": 800000, + "deprecation_date": "2026-09-30" }, "gemini/gemini-3.1-pro-preview": { "prompt_cache_min_tokens": 4096, @@ -26146,7 +26147,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "low/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26157,7 +26159,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "low/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26168,7 +26171,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.034, @@ -26179,7 +26183,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.05, @@ -26190,7 +26195,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.05, @@ -26201,7 +26207,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.133, @@ -26212,7 +26219,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.2, @@ -26223,7 +26231,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.2, @@ -26234,7 +26243,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.009, @@ -26245,7 +26255,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26256,7 +26267,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26267,7 +26279,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.009, @@ -26278,7 +26291,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26289,7 +26303,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26300,7 +26315,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "gpt-5": { "cache_read_input_token_cost": 1.25e-07, @@ -44050,21 +44066,24 @@ ] }, "xai/grok-3": { - "cache_read_input_token_cost": 7.5e-07, - "input_cost_per_token": 3e-06, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://x.ai/api#pricing", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": false, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-3-beta": { "cache_read_input_token_cost": 7.5e-07, @@ -44115,21 +44134,24 @@ "supports_web_search": true }, "xai/grok-3-latest": { - "cache_read_input_token_cost": 7.5e-07, - "input_cost_per_token": 3e-06, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://x.ai/api#pricing", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": false, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-3-mini": { "cache_read_input_token_cost": 7.5e-08, @@ -44236,89 +44258,99 @@ "supports_web_search": true }, "xai/grok-4": { - "input_cost_per_token": 3e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-fast-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-fast-non-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-0709": { - "input_cost_per_token": 3e-06, - "input_cost_per_token_above_128k_tokens": 6e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, - "output_cost_per_token_above_128k_tokens": 3e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-latest": { - "input_cost_per_token": 3e-06, - "input_cost_per_token_above_128k_tokens": 6e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, - "output_cost_per_token_above_128k_tokens": 3e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast": { "cache_read_input_token_cost": 5e-08, @@ -44342,16 +44374,14 @@ "supports_web_search": true }, "xai/grok-4-1-fast-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44361,19 +44391,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-reasoning-latest": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44383,19 +44414,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-non-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-non-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44404,19 +44436,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-non-reasoning-latest": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-non-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44425,7 +44458,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4.20-multi-agent-beta-0309": { "cache_read_input_token_cost": 2e-07, @@ -50432,14 +50468,13 @@ "us.xai.grok-4.6": { "input_cost_per_token": 2.2e-06, "output_cost_per_token": 6.6e-06, - "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 500000, "max_output_tokens": 500000, "max_tokens": 500000, "mode": "chat", "supports_function_calling": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true @@ -50447,14 +50482,13 @@ "global.xai.grok-4.6": { "input_cost_per_token": 2e-06, "output_cost_per_token": 6e-06, - "cache_read_input_token_cost": 5e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 500000, "max_output_tokens": 500000, "max_tokens": 500000, "mode": "chat", "supports_function_calling": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index ab1054d985b..390489a76f6 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -23096,7 +23096,8 @@ "supports_system_messages": true, "supports_video_input": true, "supports_vision": true, - "tpm": 800000 + "tpm": 800000, + "deprecation_date": "2026-09-30" }, "gemini/gemini-3.1-pro-preview": { "prompt_cache_min_tokens": 4096, @@ -26146,7 +26147,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "low/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26157,7 +26159,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "low/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26168,7 +26171,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.034, @@ -26179,7 +26183,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.05, @@ -26190,7 +26195,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "medium/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.05, @@ -26201,7 +26207,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.133, @@ -26212,7 +26219,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.2, @@ -26223,7 +26231,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "high/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.2, @@ -26234,7 +26243,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.009, @@ -26245,7 +26255,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26256,7 +26267,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "standard/1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26267,7 +26279,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1024-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.009, @@ -26278,7 +26291,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1024-x-1536/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26289,7 +26303,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "1536-x-1024/gpt-image-1.5-2025-12-16": { "input_cost_per_image": 0.013, @@ -26300,7 +26315,8 @@ "/v1/images/edits" ], "supports_vision": true, - "supports_pdf_input": true + "supports_pdf_input": true, + "deprecation_date": "2026-12-01" }, "gpt-5": { "cache_read_input_token_cost": 1.25e-07, @@ -44050,21 +44066,24 @@ ] }, "xai/grok-3": { - "cache_read_input_token_cost": 7.5e-07, - "input_cost_per_token": 3e-06, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://x.ai/api#pricing", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": false, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-3-beta": { "cache_read_input_token_cost": 7.5e-07, @@ -44115,21 +44134,24 @@ "supports_web_search": true }, "xai/grok-3-latest": { - "cache_read_input_token_cost": 7.5e-07, - "input_cost_per_token": 3e-06, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 131072, "max_output_tokens": 131072, "max_tokens": 131072, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://x.ai/api#pricing", "supports_function_calling": true, "supports_prompt_caching": true, "supports_response_schema": false, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-3-mini": { "cache_read_input_token_cost": 7.5e-08, @@ -44236,89 +44258,99 @@ "supports_web_search": true }, "xai/grok-4": { - "input_cost_per_token": 3e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-fast-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-fast-non-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-0709": { - "input_cost_per_token": 3e-06, - "input_cost_per_token_above_128k_tokens": 6e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, - "output_cost_per_token_above_128k_tokens": 3e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-latest": { - "input_cost_per_token": 3e-06, - "input_cost_per_token_above_128k_tokens": 6e-06, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 256000, "max_output_tokens": 256000, "max_tokens": 256000, "mode": "chat", - "output_cost_per_token": 1.5e-05, - "output_cost_per_token_above_128k_tokens": 3e-05, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models", "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast": { "cache_read_input_token_cost": 5e-08, @@ -44342,16 +44374,14 @@ "supports_web_search": true }, "xai/grok-4-1-fast-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44361,19 +44391,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-reasoning-latest": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44383,19 +44414,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-non-reasoning": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-non-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44404,19 +44436,20 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4-1-fast-non-reasoning-latest": { - "cache_read_input_token_cost": 5e-08, - "input_cost_per_token": 2e-07, - "input_cost_per_token_above_128k_tokens": 4e-07, + "cache_read_input_token_cost": 2e-07, + "input_cost_per_token": 1.25e-06, "litellm_provider": "xai", "max_input_tokens": 2000000.0, "max_output_tokens": 2000000.0, "max_tokens": 2000000.0, "mode": "chat", - "output_cost_per_token": 5e-07, - "output_cost_per_token_above_128k_tokens": 1e-06, + "output_cost_per_token": 2.5e-06, "source": "https://docs.x.ai/docs/models/grok-4-1-fast-non-reasoning", "supports_audio_input": true, "supports_function_calling": true, @@ -44425,7 +44458,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_web_search": true, - "deprecation_date": "2026-05-15" + "deprecation_date": "2026-05-15", + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token_above_200k_tokens": 5e-06, + "cache_read_input_token_cost_above_200k_tokens": 4e-07 }, "xai/grok-4.20-multi-agent-beta-0309": { "cache_read_input_token_cost": 2e-07, @@ -50432,14 +50468,13 @@ "us.xai.grok-4.6": { "input_cost_per_token": 2.2e-06, "output_cost_per_token": 6.6e-06, - "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 500000, "max_output_tokens": 500000, "max_tokens": 500000, "mode": "chat", "supports_function_calling": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true @@ -50447,14 +50482,13 @@ "global.xai.grok-4.6": { "input_cost_per_token": 2e-06, "output_cost_per_token": 6e-06, - "cache_read_input_token_cost": 5e-07, "litellm_provider": "bedrock_converse", "max_input_tokens": 500000, "max_output_tokens": 500000, "max_tokens": 500000, "mode": "chat", "supports_function_calling": true, - "supports_prompt_caching": true, + "supports_prompt_caching": false, "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true diff --git a/tests/test_litellm/integrations/test_anthropic_cache_control_hook.py b/tests/test_litellm/integrations/test_anthropic_cache_control_hook.py index 3a736e2a889..51634c4b723 100644 --- a/tests/test_litellm/integrations/test_anthropic_cache_control_hook.py +++ b/tests/test_litellm/integrations/test_anthropic_cache_control_hook.py @@ -1611,6 +1611,18 @@ class TestEnableAnthropicPromptCaching: monkeypatch.setattr(litellm, "enable_anthropic_prompt_caching", True) assert self._points(model="anthropic.claude-3-5-sonnet-20240620-v1:0", provider="bedrock") == [] + @pytest.mark.parametrize("model", ["us.xai.grok-4.6", "global.xai.grok-4.6"]) + def test_bedrock_grok_not_injected(self, monkeypatch, local_model_cost_map, model): + """Bedrock does not support prompt caching for Grok. The cost-map entries must + carry supports_prompt_caching=false so no breakpoints are injected; otherwise + Bedrock rejects the whole request ("You invoked an unsupported model or your + request did not allow prompt caching").""" + from litellm.utils import supports_prompt_caching + + monkeypatch.setattr(litellm, "enable_anthropic_prompt_caching", True) + assert supports_prompt_caching(model=model, custom_llm_provider="bedrock") is False + assert self._points(model=model, provider="bedrock") == [] + def test_stands_down_when_client_sent_cache_control(self, monkeypatch): monkeypatch.setattr(litellm, "enable_anthropic_prompt_caching", True) messages = [ diff --git a/tests/test_litellm/llms/xai/test_xai_cost_calculator.py b/tests/test_litellm/llms/xai/test_xai_cost_calculator.py index 55e28dff81d..94cd10437e2 100644 --- a/tests/test_litellm/llms/xai/test_xai_cost_calculator.py +++ b/tests/test_litellm/llms/xai/test_xai_cost_calculator.py @@ -127,11 +127,12 @@ class TestXAICostCalculator: prompt_cost, completion_cost = cost_per_token(model="grok-4", usage=usage) - # Expected costs for grok-4: - # Input: 10 tokens * $3e-6 = $0.00003 - # Completion: (200 + 150) tokens * $1.5e-5 = $0.00525 - expected_prompt_cost = 10 * 3e-6 - expected_completion_cost = (200 + 150) * 1.5e-5 + # grok-4 was retired on 2026-05-15 and now redirects to grok-4.3, so it bills + # at grok-4.3's rates: + # Input: 10 tokens * $1.25e-6 + # Completion: (200 + 150) tokens * $2.5e-6 + expected_prompt_cost = 10 * 1.25e-6 + expected_completion_cost = (200 + 150) * 2.5e-6 assert math.isclose(prompt_cost, expected_prompt_cost, rel_tol=1e-10) assert math.isclose(completion_cost, expected_completion_cost, rel_tol=1e-10) @@ -193,7 +194,7 @@ class TestXAICostCalculator: def test_tiered_pricing_above_128k_tokens(self): """Test tiered pricing for tokens above 128k.""" - # Test with grok-4-fast-reasoning which has tiered pricing + # Test with grok-4-1-fast, which still carries a 128k tier usage = Usage( prompt_tokens=150000, # Above 128k threshold completion_tokens=100000, # Above 128k threshold @@ -208,10 +209,10 @@ class TestXAICostCalculator: ) prompt_cost, completion_cost = cost_per_token( - model="xai/grok-4-fast-reasoning", usage=usage + model="xai/grok-4-1-fast", usage=usage ) - # Expected costs for grok-4-fast-reasoning with tiered pricing: + # Expected costs for grok-4-1-fast with tiered pricing: # Input: 150000 tokens * $0.4e-6 (ALL tokens at tiered rate since input > 128k) = $0.06 # Completion: (100000 + 50000) tokens * $1e-6 (tiered rate since input > 128k) = $0.15 expected_prompt_cost = 150000 * 0.4e-6 @@ -222,7 +223,7 @@ class TestXAICostCalculator: def test_tiered_pricing_below_128k_tokens(self): """Test that regular pricing is used for tokens below 128k threshold.""" - # Test with grok-4-fast-reasoning which has tiered pricing + # Test with grok-4-1-fast, which still carries a 128k tier usage = Usage( prompt_tokens=100000, # Below 128k threshold completion_tokens=50000, @@ -237,10 +238,10 @@ class TestXAICostCalculator: ) prompt_cost, completion_cost = cost_per_token( - model="xai/grok-4-fast-reasoning", usage=usage + model="xai/grok-4-1-fast", usage=usage ) - # Expected costs for grok-4-fast-reasoning with regular pricing: + # Expected costs for grok-4-1-fast with regular pricing: # Input: 100000 tokens * $0.2e-6 (regular rate) = $0.02 # Completion: (50000 + 10000) tokens * $0.5e-6 (regular rate) = $0.03 expected_prompt_cost = 100000 * 0.2e-6 @@ -252,9 +253,9 @@ class TestXAICostCalculator: def test_tiered_pricing_grok_4_latest(self): """Test tiered pricing for grok-4-latest model.""" usage = Usage( - prompt_tokens=200000, # Above 128k threshold + prompt_tokens=250000, # Above the 200k threshold completion_tokens=100000, - total_tokens=350000, + total_tokens=400000, completion_tokens_details=CompletionTokensDetailsWrapper( accepted_prediction_tokens=0, audio_tokens=0, @@ -268,11 +269,11 @@ class TestXAICostCalculator: model="xai/grok-4-latest", usage=usage ) - # Expected costs for grok-4-latest with tiered pricing: - # Input: 200000 tokens * $6e-6 (ALL tokens at tiered rate since input > 128k) = $1.2 - # Completion: (100000 + 50000) tokens * $30e-6 (tiered rate since input > 128k) = $4.5 - expected_prompt_cost = 200000 * 6e-6 - expected_completion_cost = (100000 + 50000) * 30e-6 + # grok-4-latest redirects to grok-4.3, which tiers at 200k rather than 128k: + # Input: 250000 tokens * $2.5e-6 (ALL tokens at tiered rate since input > 200k) + # Completion: (100000 + 50000) tokens * $5e-6 (tiered rate since input > 200k) + expected_prompt_cost = 250000 * 2.5e-6 + expected_completion_cost = (100000 + 50000) * 5e-6 assert math.isclose(prompt_cost, expected_prompt_cost, rel_tol=1e-10) assert math.isclose(completion_cost, expected_completion_cost, rel_tol=1e-10) @@ -293,10 +294,10 @@ class TestXAICostCalculator: ) prompt_cost, completion_cost = cost_per_token( - model="xai/grok-4-fast-reasoning", usage=usage + model="xai/grok-4-1-fast", usage=usage ) - # Expected costs for grok-4-fast-reasoning: + # Expected costs for grok-4-1-fast: # Input: 150000 tokens * $0.4e-6 (ALL tokens at tiered rate since input > 128k) = $0.06 # Completion: (50000 + 10000) tokens * $1e-6 (tiered rate since input > 128k) = $0.06 expected_prompt_cost = 150000 * 0.4e-6 diff --git a/tests/test_litellm/llms/xai/test_xai_redirected_slug_pricing.py b/tests/test_litellm/llms/xai/test_xai_redirected_slug_pricing.py new file mode 100644 index 00000000000..da1c3ed1ed8 --- /dev/null +++ b/tests/test_litellm/llms/xai/test_xai_redirected_slug_pricing.py @@ -0,0 +1,133 @@ +""" +xAI retired eight slugs on 2026-05-15 but kept them resolvable: chat slugs redirect to +grok-4.3 and bill at grok-4.3's rates, while the grok-code-fast slugs are aliases of +grok-build-0.1 and bill at its rates, so the registry must price them that way or spend +tracking is wrong. + +https://docs.x.ai/developers/migration/may-15-retirement +https://docs.x.ai/developers/models/grok-build-0.1 +""" + +from __future__ import annotations + +import json +from pathlib import Path + +import pytest + +REPO_ROOT = Path(__file__).parents[4] +PRICES_PATH = REPO_ROOT / "model_prices_and_context_window.json" +BACKUP_PRICES_PATH = REPO_ROOT / "litellm" / "model_prices_and_context_window_backup.json" +MAP_PATHS = (PRICES_PATH, BACKUP_PRICES_PATH) + +REDIRECT_TARGET = "xai/grok-4.3" +REDIRECTED_SLUGS = ( + "xai/grok-3", + "xai/grok-3-latest", + "xai/grok-4", + "xai/grok-4-0709", + "xai/grok-4-1-fast-non-reasoning", + "xai/grok-4-1-fast-non-reasoning-latest", + "xai/grok-4-1-fast-reasoning", + "xai/grok-4-1-fast-reasoning-latest", + "xai/grok-4-fast-non-reasoning", + "xai/grok-4-fast-reasoning", + "xai/grok-4-latest", +) +CODE_REDIRECT_TARGET = "xai/grok-build-0.1" +CODE_SLUGS = ( + "xai/grok-code-fast", + "xai/grok-code-fast-1", + "xai/grok-code-fast-1-0825", +) +RETIREMENT_DATE = "2026-05-15" + +BASE_COST_FIELDS = ("input_cost_per_token", "output_cost_per_token", "cache_read_input_token_cost") +TIER_COST_FIELDS = ( + "input_cost_per_token_above_200k_tokens", + "output_cost_per_token_above_200k_tokens", + "cache_read_input_token_cost_above_200k_tokens", +) +STALE_TIER_FIELDS = ( + "input_cost_per_token_above_128k_tokens", + "output_cost_per_token_above_128k_tokens", + "cache_read_input_token_cost_above_128k_tokens", +) + + +@pytest.fixture(scope="module", params=[p.name for p in MAP_PATHS]) +def cost_map(request: pytest.FixtureRequest) -> dict: + path = next(p for p in MAP_PATHS if p.name == request.param) + return json.loads(path.read_text(encoding="utf-8")) + + +@pytest.mark.parametrize("slug", REDIRECTED_SLUGS) +def test_redirected_slug_bills_at_the_target_rate(cost_map: dict, slug: str): + target = cost_map[REDIRECT_TARGET] + entry = cost_map[slug] + for field in BASE_COST_FIELDS: + assert entry[field] == target[field], field + + +@pytest.mark.parametrize("slug", CODE_SLUGS) +def test_code_slug_bills_at_grok_build_rate(cost_map: dict, slug: str): + """grok-code-fast* are aliases of grok-build-0.1, not grok-4.3 redirects.""" + target = cost_map[CODE_REDIRECT_TARGET] + entry = cost_map[slug] + for field in (*BASE_COST_FIELDS, *TIER_COST_FIELDS): + assert entry[field] == target[field], field + + +@pytest.mark.parametrize("slug", (*REDIRECTED_SLUGS, *CODE_SLUGS)) +def test_redirected_slug_keeps_its_retirement_date(cost_map: dict, slug: str): + assert cost_map[slug]["deprecation_date"] == RETIREMENT_DATE + + +@pytest.mark.parametrize("slug", REDIRECTED_SLUGS) +def test_no_slug_keeps_the_superseded_128k_tier(cost_map: dict, slug: str): + """The 128k tier belonged to the retired model; grok-4.3 tiers at 200k.""" + for field in STALE_TIER_FIELDS: + assert field not in cost_map[slug], field + + +@pytest.mark.parametrize("slug", REDIRECTED_SLUGS) +def test_redirected_slug_carries_the_target_tier_rates(cost_map: dict, slug: str): + """The request executes as grok-4.3, so it is tiered at grok-4.3's 200k boundary.""" + target = cost_map[REDIRECT_TARGET] + entry = cost_map[slug] + for field in TIER_COST_FIELDS: + assert entry[field] == target[field], field + + +def test_a_live_xai_model_is_untouched(cost_map: dict): + """Guard against the repricing leaking onto models xAI still serves directly.""" + assert cost_map["xai/grok-4.6"]["input_cost_per_token"] != cost_map[REDIRECT_TARGET]["input_cost_per_token"] + assert "deprecation_date" not in cost_map["xai/grok-4.6"] + + +def test_both_cost_maps_agree_on_the_redirected_slugs(): + prices = json.loads(PRICES_PATH.read_text(encoding="utf-8")) + backup = json.loads(BACKUP_PRICES_PATH.read_text(encoding="utf-8")) + for slug in (*REDIRECTED_SLUGS, *CODE_SLUGS, REDIRECT_TARGET, CODE_REDIRECT_TARGET): + assert prices[slug] == backup[slug], slug + + +def test_every_retired_chat_slug_is_covered(cost_map: dict): + """The list above must stay in step with what the registry marks retired.""" + marked = { + key + for key, entry in cost_map.items() + if isinstance(entry, dict) + and entry.get("litellm_provider") == "xai" + and entry.get("deprecation_date") == RETIREMENT_DATE + and entry.get("mode") == "chat" + } + assert marked == {*REDIRECTED_SLUGS, *CODE_SLUGS} + + +def test_a_live_tiered_model_keeps_its_own_rates(cost_map: dict): + """grok-4-1-fast is not retired, so it must keep its 128k tier and its own prices.""" + entry = cost_map["xai/grok-4-1-fast"] + assert "deprecation_date" not in entry + assert entry["input_cost_per_token_above_128k_tokens"] == 4e-07 + assert entry["input_cost_per_token"] == 2e-07