From bb8e0cd3e21aeea2cf23cbb3de28e37a1a492bfc Mon Sep 17 00:00:00 2001 From: Sameer Kankute Date: Tue, 17 Mar 2026 22:47:45 +0530 Subject: [PATCH] Update 5.4 family values correctly --- ...odel_prices_and_context_window_backup.json | 32 +++++-------------- model_prices_and_context_window.json | 32 +++++-------------- 2 files changed, 16 insertions(+), 48 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 4a627d4aa1..0be7aefa36 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -18749,27 +18749,19 @@ }, "gpt-5.4-mini": { "cache_read_input_token_cost": 7.5e-08, - "cache_read_input_token_cost_above_272k_tokens": 1.5e-07, - "cache_read_input_token_cost_flex": 3.9e-08, - "cache_read_input_token_cost_priority": 1.5e-07, - "cache_read_input_token_cost_above_272k_tokens_priority": 3e-07, + "cache_read_input_token_cost_flex": 3.75e-08, + "cache_read_input_token_cost_batches": 3.8e-08, "input_cost_per_token": 7.5e-07, - "input_cost_per_token_above_272k_tokens": 1.5e-06, "input_cost_per_token_flex": 3.75e-07, "input_cost_per_token_batches": 3.75e-07, - "input_cost_per_token_priority": 1.5e-06, - "input_cost_per_token_above_272k_tokens_priority": 3e-06, "litellm_provider": "openai", - "max_input_tokens": 1050000, + "max_input_tokens": 272000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 4.5e-06, - "output_cost_per_token_above_272k_tokens": 6.75e-06, "output_cost_per_token_flex": 2.25e-06, "output_cost_per_token_batches": 2.25e-06, - "output_cost_per_token_priority": 6.75e-06, - "output_cost_per_token_above_272k_tokens_priority": 1.0125e-05, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -18798,27 +18790,19 @@ }, "gpt-5.4-nano": { "cache_read_input_token_cost": 2e-08, - "cache_read_input_token_cost_above_272k_tokens": 4e-08, - "cache_read_input_token_cost_flex": 1.04e-08, - "cache_read_input_token_cost_priority": 4e-08, - "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08, + "cache_read_input_token_cost_flex": 1e-08, + "cache_read_input_token_cost_batches": 1e-08, "input_cost_per_token": 2e-07, - "input_cost_per_token_above_272k_tokens": 4e-07, "input_cost_per_token_flex": 1e-07, "input_cost_per_token_batches": 1e-07, - "input_cost_per_token_priority": 4e-07, - "input_cost_per_token_above_272k_tokens_priority": 8e-07, "litellm_provider": "openai", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 1.25e-06, - "output_cost_per_token_above_272k_tokens": 1.875e-06, "output_cost_per_token_flex": 6.25e-07, "output_cost_per_token_batches": 6.25e-07, - "output_cost_per_token_priority": 1.875e-06, - "output_cost_per_token_above_272k_tokens_priority": 2.8125e-06, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 4a627d4aa1..0be7aefa36 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -18749,27 +18749,19 @@ }, "gpt-5.4-mini": { "cache_read_input_token_cost": 7.5e-08, - "cache_read_input_token_cost_above_272k_tokens": 1.5e-07, - "cache_read_input_token_cost_flex": 3.9e-08, - "cache_read_input_token_cost_priority": 1.5e-07, - "cache_read_input_token_cost_above_272k_tokens_priority": 3e-07, + "cache_read_input_token_cost_flex": 3.75e-08, + "cache_read_input_token_cost_batches": 3.8e-08, "input_cost_per_token": 7.5e-07, - "input_cost_per_token_above_272k_tokens": 1.5e-06, "input_cost_per_token_flex": 3.75e-07, "input_cost_per_token_batches": 3.75e-07, - "input_cost_per_token_priority": 1.5e-06, - "input_cost_per_token_above_272k_tokens_priority": 3e-06, "litellm_provider": "openai", - "max_input_tokens": 1050000, + "max_input_tokens": 272000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 4.5e-06, - "output_cost_per_token_above_272k_tokens": 6.75e-06, "output_cost_per_token_flex": 2.25e-06, "output_cost_per_token_batches": 2.25e-06, - "output_cost_per_token_priority": 6.75e-06, - "output_cost_per_token_above_272k_tokens_priority": 1.0125e-05, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -18798,27 +18790,19 @@ }, "gpt-5.4-nano": { "cache_read_input_token_cost": 2e-08, - "cache_read_input_token_cost_above_272k_tokens": 4e-08, - "cache_read_input_token_cost_flex": 1.04e-08, - "cache_read_input_token_cost_priority": 4e-08, - "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08, + "cache_read_input_token_cost_flex": 1e-08, + "cache_read_input_token_cost_batches": 1e-08, "input_cost_per_token": 2e-07, - "input_cost_per_token_above_272k_tokens": 4e-07, "input_cost_per_token_flex": 1e-07, "input_cost_per_token_batches": 1e-07, - "input_cost_per_token_priority": 4e-07, - "input_cost_per_token_above_272k_tokens_priority": 8e-07, "litellm_provider": "openai", - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "max_tokens": 16384, + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, "mode": "chat", "output_cost_per_token": 1.25e-06, - "output_cost_per_token_above_272k_tokens": 1.875e-06, "output_cost_per_token_flex": 6.25e-07, "output_cost_per_token_batches": 6.25e-07, - "output_cost_per_token_priority": 1.875e-06, - "output_cost_per_token_above_272k_tokens_priority": 2.8125e-06, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch",