From e4a557d95fe74e3a08a79ce20c3fe83ad5a5888f Mon Sep 17 00:00:00 2001 From: Cesar Garcia <128240629+Chesars@users.noreply.github.com> Date: Tue, 27 Jan 2026 02:00:35 -0300 Subject: [PATCH] fix(xai): correct cached token cost calculation for xAI models (#19772) * fix(azure): use generic cost calculator for audio token pricing Azure audio models were charging audio output tokens at the text token rate instead of the correct audio token rate. This resulted in costs being ~6.65x lower than expected. The fix replaces Azure's custom cost calculation logic with the generic cost calculator that properly handles text, audio, cached, reasoning, and image tokens. Fixes #19764 * fix(xai): correct cached token cost calculation for xAI models - Fix double-counting issue where xAI reports text_tokens = prompt_tokens (including cached), causing tokens to be charged twice - Add cache_read_input_token_cost to xAI grok-3 and grok-3-mini model variants - Detection: when text_tokens + cached_tokens > prompt_tokens, recalculate text_tokens = prompt_tokens - cached_tokens xAI pricing (25% of input for cached): - grok-3 variants: $0.75/M cached (input $3/M) - grok-3-mini variants: $0.075/M cached (input $0.30/M) --- .../litellm_core_utils/llm_cost_calc/utils.py | 24 +++++++++++++++---- ...odel_prices_and_context_window_backup.json | 11 +++++++++ model_prices_and_context_window.json | 11 +++++++++ 3 files changed, 41 insertions(+), 5 deletions(-) diff --git a/litellm/litellm_core_utils/llm_cost_calc/utils.py b/litellm/litellm_core_utils/llm_cost_calc/utils.py index 785976ed319..642cbbb7922 100644 --- a/litellm/litellm_core_utils/llm_cost_calc/utils.py +++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py @@ -566,14 +566,28 @@ def generic_cost_per_token( # noqa: PLR0915 if usage.prompt_tokens_details: prompt_tokens_details = _parse_prompt_tokens_details(usage) - ## EDGE CASE - text tokens not set inside PromptTokensDetails + ## EDGE CASE - text tokens not set or includes cached tokens (double-counting) + ## Some providers (like xAI) report text_tokens = prompt_tokens (including cached) + ## We detect this when: text_tokens + cached_tokens + other > prompt_tokens + ## Ref: https://github.com/BerriAI/litellm/issues/19680, #14874, #14875 - if prompt_tokens_details["text_tokens"] == 0: + cache_hit = prompt_tokens_details["cache_hit_tokens"] + text_tokens = prompt_tokens_details["text_tokens"] + audio_tokens = prompt_tokens_details["audio_tokens"] + cache_creation = prompt_tokens_details["cache_creation_tokens"] + image_tokens = prompt_tokens_details["image_tokens"] + + # Check for double-counting: sum of details > prompt_tokens means overlap + total_details = text_tokens + cache_hit + audio_tokens + cache_creation + image_tokens + has_double_counting = cache_hit > 0 and total_details > usage.prompt_tokens + + if text_tokens == 0 or has_double_counting: text_tokens = ( usage.prompt_tokens - - prompt_tokens_details["cache_hit_tokens"] - - prompt_tokens_details["audio_tokens"] - - prompt_tokens_details["cache_creation_tokens"] + - cache_hit + - audio_tokens + - cache_creation + - image_tokens ) prompt_tokens_details["text_tokens"] = text_tokens diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index d958ea4503a..cf6b0e7e3f2 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -30382,6 +30382,7 @@ "supports_web_search": true }, "xai/grok-3": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30396,6 +30397,7 @@ "supports_web_search": true }, "xai/grok-3-beta": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30410,6 +30412,7 @@ "supports_web_search": true }, "xai/grok-3-fast-beta": { + "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 5e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30424,6 +30427,7 @@ "supports_web_search": true }, "xai/grok-3-fast-latest": { + "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 5e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30438,6 +30442,7 @@ "supports_web_search": true }, "xai/grok-3-latest": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30452,6 +30457,7 @@ "supports_web_search": true }, "xai/grok-3-mini": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30467,6 +30473,7 @@ "supports_web_search": true }, "xai/grok-3-mini-beta": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30482,6 +30489,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30497,6 +30505,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast-beta": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30512,6 +30521,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast-latest": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30527,6 +30537,7 @@ "supports_web_search": true }, "xai/grok-3-mini-latest": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 244a679bc27..d0a3bec9009 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -30384,6 +30384,7 @@ "supports_web_search": true }, "xai/grok-3": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30398,6 +30399,7 @@ "supports_web_search": true }, "xai/grok-3-beta": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30412,6 +30414,7 @@ "supports_web_search": true }, "xai/grok-3-fast-beta": { + "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 5e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30426,6 +30429,7 @@ "supports_web_search": true }, "xai/grok-3-fast-latest": { + "cache_read_input_token_cost": 1.25e-06, "input_cost_per_token": 5e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30440,6 +30444,7 @@ "supports_web_search": true }, "xai/grok-3-latest": { + "cache_read_input_token_cost": 7.5e-07, "input_cost_per_token": 3e-06, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30454,6 +30459,7 @@ "supports_web_search": true }, "xai/grok-3-mini": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30469,6 +30475,7 @@ "supports_web_search": true }, "xai/grok-3-mini-beta": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30484,6 +30491,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30499,6 +30507,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast-beta": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30514,6 +30523,7 @@ "supports_web_search": true }, "xai/grok-3-mini-fast-latest": { + "cache_read_input_token_cost": 1.5e-07, "input_cost_per_token": 6e-07, "litellm_provider": "xai", "max_input_tokens": 131072, @@ -30529,6 +30539,7 @@ "supports_web_search": true }, "xai/grok-3-mini-latest": { + "cache_read_input_token_cost": 7.5e-08, "input_cost_per_token": 3e-07, "litellm_provider": "xai", "max_input_tokens": 131072,