From 1bbfadad5b27de6bab68c1adee96ec5904ecbf5e Mon Sep 17 00:00:00 2001 From: Jeremy Luo Date: Mon, 20 Jul 2026 15:17:37 +0800 Subject: [PATCH] feat(bedrock): add OpenAI GPT-5.6 (Sol/Terra/Luna) support via bedrock-mantle endpoint - Add GPT-5.6 Sol/Terra/Luna as selectable models on the bedrock-mantle Responses API path (builds on #3240 for GPT-5.5/5.4) - Add reasoning_effort parameter (none/low/medium/high/xhigh/max), forwarded to the Responses API reasoning.effort - Do not forward top_p on the mantle path: rejected by GPT-5.5/5.6 and incompatible with reasoning even on GPT-5.4 - Exclude the GPT-5.6 models from cross-region inference (In-Region only) - Bump plugin version to 0.0.75 --- models/bedrock/manifest.yaml | 2 +- models/bedrock/models/llm/llm.py | 32 ++++++++++- .../model_configurations/gpt-5-6-luna.yaml | 55 +++++++++++++++++++ .../llm/model_configurations/gpt-5-6-sol.yaml | 55 +++++++++++++++++++ .../model_configurations/gpt-5-6-terra.yaml | 55 +++++++++++++++++++ models/bedrock/models/llm/model_ids.py | 6 ++ models/bedrock/models/llm/openai.yaml | 20 +++++++ 7 files changed, 221 insertions(+), 4 deletions(-) create mode 100644 models/bedrock/models/llm/model_configurations/gpt-5-6-luna.yaml create mode 100644 models/bedrock/models/llm/model_configurations/gpt-5-6-sol.yaml create mode 100644 models/bedrock/models/llm/model_configurations/gpt-5-6-terra.yaml diff --git a/models/bedrock/manifest.yaml b/models/bedrock/manifest.yaml index 026ff8bf7..89c5ae6fd 100644 --- a/models/bedrock/manifest.yaml +++ b/models/bedrock/manifest.yaml @@ -1,4 +1,4 @@ -version: 0.0.74 +version: 0.0.75 type: plugin author: langgenius name: bedrock diff --git a/models/bedrock/models/llm/llm.py b/models/bedrock/models/llm/llm.py index 281e6f9b5..6b84b824f 100644 --- a/models/bedrock/models/llm/llm.py +++ b/models/bedrock/models/llm/llm.py @@ -106,6 +106,9 @@ class BedrockLargeLanguageModel(LargeLanguageModel): # Models that require the bedrock-mantle endpoint with the OpenAI Responses API. # These do NOT go through the standard Converse API path. _BEDROCK_MANTLE_MODEL_IDS: frozenset = frozenset({ + "openai.gpt-5.6-sol", + "openai.gpt-5.6-terra", + "openai.gpt-5.6-luna", "openai.gpt-5.5", "openai.gpt-5.4", }) @@ -1856,6 +1859,9 @@ def _get_model_specific_pricing(self, model: str, model_name: str, model_schemas # Create model name mapping for individual model files model_name_mapping = { # OpenAI GPT-5.x models (bedrock-mantle endpoint) + 'GPT-5.6 Sol': 'gpt-5-6-sol', + 'GPT-5.6 Terra': 'gpt-5-6-terra', + 'GPT-5.6 Luna': 'gpt-5-6-luna', 'GPT-5.5': 'gpt-5-5', 'GPT-5.4': 'gpt-5-4', # Claude models @@ -2133,12 +2139,32 @@ def _generate_with_responses_api( params["max_output_tokens"] = model_parameters["max_tokens"] if "temperature" in model_parameters: params["temperature"] = model_parameters["temperature"] - if "top_p" in model_parameters: - params["top_p"] = model_parameters["top_p"] + # NOTE: top_p is intentionally NOT forwarded on the bedrock-mantle path. + # GPT-5.5 / GPT-5.6 reject top_p outright (400 unsupported_parameter), and + # even GPT-5.4 rejects it once reasoning is active (which it is by default + # here). These are reasoning models driven by reasoning_effort, so top_p + # does not apply. It remains in openai.yaml only for the GPT OSS Converse + # models, which are handled on a separate code path. + # GPT-5.6/5.5/5.4 reasoning models: pass reasoning effort through the + # Responses API. Effort values: none, low, medium, high, xhigh, max. + reasoning_effort = model_parameters.get("reasoning_effort") + if reasoning_effort: + params["reasoning"] = {"effort": reasoning_effort} # Store model_name for pricing calculation, deriving it from model_id if not set. credentials_for_pricing = credentials.copy() - resolved_model_name = model_name or ("GPT-5.5" if "gpt-5.5" in model_id else "GPT-5.4") + if model_name: + resolved_model_name = model_name + elif "gpt-5.6-sol" in model_id: + resolved_model_name = "GPT-5.6 Sol" + elif "gpt-5.6-terra" in model_id: + resolved_model_name = "GPT-5.6 Terra" + elif "gpt-5.6-luna" in model_id: + resolved_model_name = "GPT-5.6 Luna" + elif "gpt-5.5" in model_id: + resolved_model_name = "GPT-5.5" + else: + resolved_model_name = "GPT-5.4" credentials_for_pricing["model_parameters"] = { **credentials_for_pricing.get("model_parameters", {}), "model_name": resolved_model_name, diff --git a/models/bedrock/models/llm/model_configurations/gpt-5-6-luna.yaml b/models/bedrock/models/llm/model_configurations/gpt-5-6-luna.yaml new file mode 100644 index 000000000..06e5f036a --- /dev/null +++ b/models/bedrock/models/llm/model_configurations/gpt-5-6-luna.yaml @@ -0,0 +1,55 @@ +model: gpt-5-6-luna +label: + en_US: GPT-5.6 Luna +icon: icon_s_en.svg +model_type: llm +features: + - vision +model_properties: + mode: chat + context_size: 272000 +parameter_rules: + - name: max_tokens + use_template: max_tokens + required: true + label: + zh_Hans: 最大token数 + en_US: Max Tokens + type: int + default: 4096 + min: 1 + max: 272000 + help: + zh_Hans: 停止前生成的最大令牌数。 + en_US: The maximum number of tokens to generate before stopping. + - name: temperature + use_template: temperature + required: false + label: + zh_Hans: 模型温度 + en_US: Model Temperature + type: float + default: 1 + min: 0.0 + max: 1.0 + help: + zh_Hans: 生成内容的随机性。 + en_US: The amount of randomness injected into the response. + - name: top_p + use_template: top_p + label: + zh_Hans: Top P + en_US: Top P + required: false + type: float + default: 0.999 + min: 0.000 + max: 1.000 + help: + zh_Hans: 在核采样中的概率阈值。 + en_US: The probability threshold in nucleus sampling. +pricing: + input: '0.0011' + output: '0.0066' + unit: '0.001' + currency: USD diff --git a/models/bedrock/models/llm/model_configurations/gpt-5-6-sol.yaml b/models/bedrock/models/llm/model_configurations/gpt-5-6-sol.yaml new file mode 100644 index 000000000..e10142cf2 --- /dev/null +++ b/models/bedrock/models/llm/model_configurations/gpt-5-6-sol.yaml @@ -0,0 +1,55 @@ +model: gpt-5-6-sol +label: + en_US: GPT-5.6 Sol +icon: icon_s_en.svg +model_type: llm +features: + - vision +model_properties: + mode: chat + context_size: 272000 +parameter_rules: + - name: max_tokens + use_template: max_tokens + required: true + label: + zh_Hans: 最大token数 + en_US: Max Tokens + type: int + default: 4096 + min: 1 + max: 272000 + help: + zh_Hans: 停止前生成的最大令牌数。 + en_US: The maximum number of tokens to generate before stopping. + - name: temperature + use_template: temperature + required: false + label: + zh_Hans: 模型温度 + en_US: Model Temperature + type: float + default: 1 + min: 0.0 + max: 1.0 + help: + zh_Hans: 生成内容的随机性。 + en_US: The amount of randomness injected into the response. + - name: top_p + use_template: top_p + label: + zh_Hans: Top P + en_US: Top P + required: false + type: float + default: 0.999 + min: 0.000 + max: 1.000 + help: + zh_Hans: 在核采样中的概率阈值。 + en_US: The probability threshold in nucleus sampling. +pricing: + input: '0.0055' + output: '0.033' + unit: '0.001' + currency: USD diff --git a/models/bedrock/models/llm/model_configurations/gpt-5-6-terra.yaml b/models/bedrock/models/llm/model_configurations/gpt-5-6-terra.yaml new file mode 100644 index 000000000..682544ec3 --- /dev/null +++ b/models/bedrock/models/llm/model_configurations/gpt-5-6-terra.yaml @@ -0,0 +1,55 @@ +model: gpt-5-6-terra +label: + en_US: GPT-5.6 Terra +icon: icon_s_en.svg +model_type: llm +features: + - vision +model_properties: + mode: chat + context_size: 272000 +parameter_rules: + - name: max_tokens + use_template: max_tokens + required: true + label: + zh_Hans: 最大token数 + en_US: Max Tokens + type: int + default: 4096 + min: 1 + max: 272000 + help: + zh_Hans: 停止前生成的最大令牌数。 + en_US: The maximum number of tokens to generate before stopping. + - name: temperature + use_template: temperature + required: false + label: + zh_Hans: 模型温度 + en_US: Model Temperature + type: float + default: 1 + min: 0.0 + max: 1.0 + help: + zh_Hans: 生成内容的随机性。 + en_US: The amount of randomness injected into the response. + - name: top_p + use_template: top_p + label: + zh_Hans: Top P + en_US: Top P + required: false + type: float + default: 0.999 + min: 0.000 + max: 1.000 + help: + zh_Hans: 在核采样中的概率阈值。 + en_US: The probability threshold in nucleus sampling. +pricing: + input: '0.00275' + output: '0.0165' + unit: '0.001' + currency: USD diff --git a/models/bedrock/models/llm/model_ids.py b/models/bedrock/models/llm/model_ids.py index 9c2de31aa..fa9bf5e06 100644 --- a/models/bedrock/models/llm/model_ids.py +++ b/models/bedrock/models/llm/model_ids.py @@ -74,6 +74,9 @@ 'Qwen3 Coder 30B': 'qwen.qwen3-coder-30b-a3b-v1:0' }, 'openai': { + 'GPT-5.6 Sol': 'openai.gpt-5.6-sol', + 'GPT-5.6 Terra': 'openai.gpt-5.6-terra', + 'GPT-5.6 Luna': 'openai.gpt-5.6-luna', 'GPT-5.5': 'openai.gpt-5.5', 'GPT-5.4': 'openai.gpt-5.4', 'GPT OSS 120B': 'openai.gpt-oss-120b-1:0', @@ -92,6 +95,9 @@ def is_support_cross_region(model_id): "openai.gpt-oss-20b-1:0", "openai.gpt-5.5", "openai.gpt-5.4", + "openai.gpt-5.6-sol", + "openai.gpt-5.6-terra", + "openai.gpt-5.6-luna", ] return model_id not in unsupport_model_list diff --git a/models/bedrock/models/llm/openai.yaml b/models/bedrock/models/llm/openai.yaml index 457f50150..52d28d235 100644 --- a/models/bedrock/models/llm/openai.yaml +++ b/models/bedrock/models/llm/openai.yaml @@ -22,6 +22,9 @@ parameter_rules: required: true default: GPT OSS 20B options: + - GPT-5.6 Sol + - GPT-5.6 Terra + - GPT-5.6 Luna - GPT-5.5 - GPT-5.4 - GPT OSS 120B @@ -79,6 +82,23 @@ parameter_rules: help: zh_Hans: 在核采样中的概率阈值。 en_US: The probability threshold in nucleus sampling. + - name: reasoning_effort + label: + zh_Hans: 推理努力程度 + en_US: Reasoning Effort + type: string + required: false + default: medium + options: + - none + - low + - medium + - high + - xhigh + - max + help: + zh_Hans: 仅对 GPT-5.6 / GPT-5.5 / GPT-5.4（bedrock-mantle Responses API）生效，控制推理投入程度，默认 medium。GPT OSS 等模型会忽略此参数。 + en_US: Applies only to GPT-5.6 / GPT-5.5 / GPT-5.4 (bedrock-mantle Responses API) to control reasoning effort, defaulting to medium. Ignored by other models such as GPT OSS. - name: response_format use_template: response_format pricing: