From e60dbc96ada45c093209a948ea98934cc04da648 Mon Sep 17 00:00:00 2001 From: kusumakarb Date: Thu, 17 Sep 2026 15:03:33 +0530 Subject: [PATCH] fix(bedrock_mantle): price gpt-5.6-sol at its promotional rates AWS reduced GPT-5.6 Sol pricing on 2026-08-21 (20% off input, 33.3% off output), available at least through 2026-11-21. The reduction was applied to the bedrock_converse keys (us.openai.gpt-5.6-sol, global.openai.gpt-5.6-sol) and to the openai key, but the bedrock_mantle/openai.gpt-5.6-sol key still carried the pre-promotion rates, so mantle callers saw reported cost run ~24-43% above what AWS billed, depending on their input/output mix. bedrock-mantle serves Sol In-Region only, and the model card prices In-Region and Geo CRIS identically, so the mantle key now matches the existing us. converse entry exactly: input 5.5e-06 -> 4.4e-06 ($4.40 / 1M) cache write (30m) 6.875e-06 -> 5.5e-06 ($5.50 / 1M) cache read 5.5e-07 -> 4.4e-07 ($0.44 / 1M) output 3.3e-05 -> 2.2e-05 ($22.00 / 1M) plus the matching >272k long-context tier. Terra and Luna were already correct and are unchanged. Ref: https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-sol.html Ref: https://aws.amazon.com/about-aws/whats-new/2026/08/bedrock-openai-gpt-56-sol-reduced-pricing/ Co-Authored-By: Claude Opus 5 (1M context) --- ...odel_prices_and_context_window_backup.json | 16 +++++++------- model_prices_and_context_window.json | 16 +++++++------- ...bedrock_mantle_responses_transformation.py | 21 +++++++++++++++++++ 3 files changed, 37 insertions(+), 16 deletions(-) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 355e2b90e96..3de36d0daaa 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -57368,14 +57368,14 @@ "supports_tool_choice": true }, "bedrock_mantle/openai.gpt-5.6-sol": { - "input_cost_per_token": 5.5e-06, - "input_cost_per_token_above_272k_tokens": 1.1e-05, - "cache_creation_input_token_cost": 6.875e-06, - "cache_creation_input_token_cost_above_272k_tokens": 1.375e-05, - "cache_read_input_token_cost": 5.5e-07, - "cache_read_input_token_cost_above_272k_tokens": 1.1e-06, - "output_cost_per_token": 3.3e-05, - "output_cost_per_token_above_272k_tokens": 4.95e-05, + "input_cost_per_token": 4.4e-06, + "input_cost_per_token_above_272k_tokens": 8.8e-06, + "cache_creation_input_token_cost": 5.5e-06, + "cache_creation_input_token_cost_above_272k_tokens": 1.1e-05, + "cache_read_input_token_cost": 4.4e-07, + "cache_read_input_token_cost_above_272k_tokens": 8.8e-07, + "output_cost_per_token": 2.2e-05, + "output_cost_per_token_above_272k_tokens": 3.3e-05, "search_context_cost_per_query": { "search_context_size_high": 0.012, "search_context_size_low": 0.012, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 355e2b90e96..3de36d0daaa 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -57368,14 +57368,14 @@ "supports_tool_choice": true }, "bedrock_mantle/openai.gpt-5.6-sol": { - "input_cost_per_token": 5.5e-06, - "input_cost_per_token_above_272k_tokens": 1.1e-05, - "cache_creation_input_token_cost": 6.875e-06, - "cache_creation_input_token_cost_above_272k_tokens": 1.375e-05, - "cache_read_input_token_cost": 5.5e-07, - "cache_read_input_token_cost_above_272k_tokens": 1.1e-06, - "output_cost_per_token": 3.3e-05, - "output_cost_per_token_above_272k_tokens": 4.95e-05, + "input_cost_per_token": 4.4e-06, + "input_cost_per_token_above_272k_tokens": 8.8e-06, + "cache_creation_input_token_cost": 5.5e-06, + "cache_creation_input_token_cost_above_272k_tokens": 1.1e-05, + "cache_read_input_token_cost": 4.4e-07, + "cache_read_input_token_cost_above_272k_tokens": 8.8e-07, + "output_cost_per_token": 2.2e-05, + "output_cost_per_token_above_272k_tokens": 3.3e-05, "search_context_cost_per_query": { "search_context_size_high": 0.012, "search_context_size_low": 0.012, diff --git a/tests/test_litellm/llms/bedrock_mantle/test_bedrock_mantle_responses_transformation.py b/tests/test_litellm/llms/bedrock_mantle/test_bedrock_mantle_responses_transformation.py index 6758a333b35..f6487a14739 100644 --- a/tests/test_litellm/llms/bedrock_mantle/test_bedrock_mantle_responses_transformation.py +++ b/tests/test_litellm/llms/bedrock_mantle/test_bedrock_mantle_responses_transformation.py @@ -1912,6 +1912,27 @@ class TestBedrockMantleResponsesSigV4: class TestBedrockMantleResponsesPricing: + @pytest.mark.parametrize( + "model", + ["openai.gpt-5.6-sol", "openai.gpt-5.6-terra", "openai.gpt-5.6-luna"], + ) + def test_mantle_matches_in_region_converse_pricing(self, local_cost_map, model): + """bedrock-mantle serves these models In-Region only, and the AWS model + cards price In-Region and Geo CRIS identically -- so every cost field on + the mantle key must equal the `us.` converse key. A price change applied + to one namespace but not the other shows up here. + """ + mantle = litellm.model_cost[f"bedrock_mantle/{model}"] + converse = litellm.model_cost[f"us.{model}"] + + cost_fields = [k for k in converse if "cost" in k and k != "search_context_cost_per_query"] + assert cost_fields, "expected cost fields on the converse entry" + for field in cost_fields: + assert mantle.get(field) == pytest.approx(converse[field]), ( + f"{model}: {field} is {mantle.get(field)} on bedrock_mantle " + f"but {converse[field]} on us. (bedrock_converse)" + ) + def test_models_registered(self, local_cost_map): assert "bedrock_mantle/openai.gpt-5.5" in litellm.bedrock_mantle_models assert "bedrock_mantle/openai.gpt-5.4" in litellm.bedrock_mantle_models