feat(aws-bedrock): update model YAMLs [bot]

providers/aws-bedrock/ai21.jamba-1-5-large-v1:0.yaml

@@ -3,7 +3,9 @@ costs:
       output_cost_per_token: 0.000008
       region: us-east-1
 features:
+    - function_calling
     - system_messages
+    - json_output
 limits:
     context_window: 256000
     max_input_tokens: 256000
@@ -33,6 +35,7 @@ params:
       key: "n"
       maxValue: 16
       minValue: 1
+provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-jamba.html
     - https://www.ai21.com/jamba

providers/aws-bedrock/ai21.jamba-1-5-mini-v1:0.yaml

@@ -25,10 +25,13 @@ params:
       key: temperature
       maxValue: 2
       minValue: 0
+    - key: top_p
+      minValue: 0.01
     - defaultValue: 1
       key: "n"
       maxValue: 16
       minValue: 1
+provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-jamba.html
 status: active

providers/aws-bedrock/amazon.nova-2-multimodal-embeddings-v1:0.yaml

@@ -16,6 +16,7 @@ modalities:
         - embedding
 mode: embedding
 model: amazon.nova-2-multimodal-embeddings-v1:0
+provisioning: serverless
 removeParams:
     - temperature
     - top_p

providers/aws-bedrock/amazon.nova-reel-v1:0.yaml

@@ -5,6 +5,8 @@ costs:
       region: eu-west-1
     - output_cost_per_second: 0.08
       region: ap-northeast-1
+deprecationDate: "2026-03-30"
+isDeprecated: true
 modalities:
     input:
         - text

providers/aws-bedrock/amazon.nova-reel-v1:1.yaml

@@ -1,6 +1,8 @@
 costs:
     - output_cost_per_second: 0.08
       region: us-east-1
+deprecationDate: "2026-03-30"
+isDeprecated: true
 modalities:
     input:
         - text

providers/aws-bedrock/amazon.titan-embed-text-v1.yaml

@@ -21,6 +21,7 @@ modalities:
         - embedding
 mode: embedding
 model: amazon.titan-embed-text-v1
+provisioning: serverless
 removeParams:
     - max_tokens
     - temperature

providers/aws-bedrock/amazon.titan-embed-text-v2:0.yaml

@@ -66,6 +66,7 @@ modalities:
         - embedding
 mode: embedding
 model: amazon.titan-embed-text-v2:0
+provisioning: serverless
 removeParams:
     - max_tokens
     - temperature

providers/aws-bedrock/apac.amazon.nova-micro-v1:0.yaml

@@ -48,6 +48,7 @@ params:
       key: max_tokens
       maxValue: 5000
       minValue: 1
+provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html
     - https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference-support.html

providers/aws-bedrock/apac.anthropic.claude-sonnet-4-20250514-v1:0.yaml

@@ -76,13 +76,15 @@ costs:
       input_cost_per_token: 0.000003
       output_cost_per_token: 0.000015
       region: ap-east-2
+deprecationDate: "2026-06-15"
 features:
     - function_calling
     - prompt_caching
     - cache_control
     - system_messages
     - tool_choice
     - assistant_prefill
+isDeprecated: true
 limits:
     context_window: 200000
     max_input_tokens: 200000

providers/aws-bedrock/au.anthropic.claude-haiku-4-5-20251001-v1:0.yaml

@@ -38,6 +38,7 @@ params:
       key: max_tokens
       maxValue: 64000
       minValue: 1
+provisioning: serverless
 sources:
     - https://platform.claude.com/docs/en/docs/about-claude/models
     - https://platform.claude.com/docs/en/about-claude/pricing

providers/aws-bedrock/au.anthropic.claude-opus-4-6-v1.yaml

@@ -39,6 +39,7 @@ params:
       key: max_tokens
       maxValue: 128000
       minValue: 1
+provisioning: serverless
 removeParams:
     - "n"
 sources:

providers/aws-bedrock/au.anthropic.claude-sonnet-4-5-20250929-v1:0.yaml

@@ -64,6 +64,7 @@ params:
       key: max_tokens
       maxValue: 64000
       minValue: 1
+provisioning: serverless
 removeParams:
     - "n"
 sources:

providers/aws-bedrock/cohere.embed-english-v3.yaml

@@ -46,6 +46,7 @@ modalities:
         - embedding
 mode: embedding
 model: cohere.embed-english-v3
+provisioning: serverless
 removeParams:
     - temperature
     - top_p

providers/aws-bedrock/cohere.embed-multilingual-v3.yaml

@@ -46,10 +46,14 @@ modalities:
         - embedding
 mode: embedding
 model: cohere.embed-multilingual-v3
+provisioning: serverless
 removeParams:
     - stream
     - max_tokens
     - temperature
+    - top_p
+    - "n"
+    - stop
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-embed-v3.html

providers/aws-bedrock/cohere.rerank-v3-5:0.yaml

@@ -18,6 +18,7 @@ modalities:
         - text
 mode: rerank
 model: cohere.rerank-v3-5:0
+provisioning: serverless
 removeParams:
     - stream
     - temperature

providers/aws-bedrock/deepseek.v3-v1:0.yaml

@@ -27,8 +27,8 @@ features:
     - function_calling
 limits:
     context_window: 128000
-    max_output_tokens: 8000
-    max_tokens: 8000
+    max_output_tokens: 8192
+    max_tokens: 8192
 modalities:
     input:
         - text
@@ -39,8 +39,9 @@ model: deepseek.v3-v1:0
 params:
     - defaultValue: 4096
       key: max_tokens
-      maxValue: 8000
+      maxValue: 8192
       minValue: 1
+provisioning: serverless
 removeParams:
     - "n"
 sources:

providers/aws-bedrock/deepseek.v3.2.yaml

@@ -46,6 +46,7 @@ params:
       key: max_tokens
       maxValue: 8000
       minValue: 1
+provisioning: serverless
 sources:
     - https://api-docs.deepseek.com/quick_start/pricing
     - https://api-docs.deepseek.com/api/create-chat-completion

providers/aws-bedrock/eu.amazon.nova-2-lite-v1:0.yaml

@@ -1,26 +1,20 @@
 costs:
-    - cache_read_input_token_cost: 1.1e-7
-      input_cost_per_token: 4.84e-7
+    - input_cost_per_token: 4.84e-7
       output_cost_per_token: 0.000004059
       region: eu-west-3
-    - cache_read_input_token_cost: 8.5e-8
-      input_cost_per_token: 3.74e-7
+    - input_cost_per_token: 3.74e-7
       output_cost_per_token: 0.000003157
       region: eu-west-1
-    - cache_read_input_token_cost: 8.25e-8
-      input_cost_per_token: 3.63e-7
+    - input_cost_per_token: 3.63e-7
       output_cost_per_token: 0.000003205
       region: eu-south-2
-    - cache_read_input_token_cost: 8.25e-8
-      input_cost_per_token: 5.28e-7
+    - input_cost_per_token: 5.28e-7
       output_cost_per_token: 0.000004411
       region: eu-south-1
-    - cache_read_input_token_cost: 8.25e-8
-      input_cost_per_token: 3.63e-7
+    - input_cost_per_token: 3.63e-7
       output_cost_per_token: 0.000002992
       region: eu-north-1
-    - cache_read_input_token_cost: 9.75e-8
-      input_cost_per_token: 4.29e-7
+    - input_cost_per_token: 4.29e-7
       output_cost_per_token: 0.000003597
       region: eu-central-1
 features:

providers/aws-bedrock/eu.anthropic.claude-haiku-4-5-20251001-v1:0.yaml

@@ -85,6 +85,7 @@ params:
       maxValue: 128000
       minValue: 1024
       type: number
+provisioning: serverless
 removeParams:
     - top_p
 sources:

providers/aws-bedrock/eu.anthropic.claude-opus-4-5-20251101-v1:0.yaml

@@ -83,6 +83,7 @@ params:
       maxValue: 128000
       minValue: 1024
       type: number
+provisioning: serverless
 removeParams:
     - temperature
     - top_p

providers/aws-bedrock/eu.anthropic.claude-opus-4-6-v1.yaml

@@ -83,6 +83,7 @@ params:
       key: max_tokens
       maxValue: 128000
       minValue: 1
+provisioning: serverless
 sources:
     - https://platform.claude.com/docs/en/docs/about-claude/models
     - https://platform.claude.com/docs/en/about-claude/pricing

providers/aws-bedrock/eu.anthropic.claude-sonnet-4-20250514-v1:0.yaml

@@ -108,13 +108,15 @@ costs:
           output:
               - cost_per_token: 0.0000225
                 from: 200001
+deprecationDate: "2026-06-15"
 features:
     - function_calling
     - tool_choice
     - system_messages
     - cache_control
     - prompt_caching
     - assistant_prefill
+isDeprecated: true
 limits:
     context_window: 200000
     max_input_tokens: 200000

providers/aws-bedrock/eu.anthropic.claude-sonnet-4-5-20250929-v1:0.yaml

@@ -86,6 +86,7 @@ params:
       type: number
     - key: tool_choice
       type: json
+provisioning: serverless
 removeParams:
     - top_p
 sources:

providers/aws-bedrock/eu.anthropic.claude-sonnet-4-6.yaml

@@ -56,12 +56,12 @@ costs:
       output_cost_per_token_batches: 0.00000825
       region: eu-central-1
 features:
+    - assistant_prefill
+    - cache_control
     - function_calling
-    - tool_choice
     - prompt_caching
-    - cache_control
     - system_messages
-    - assistant_prefill
+    - tool_choice
 limits:
     context_window: 1000000
     max_output_tokens: 64000
@@ -80,6 +80,7 @@ params:
       key: max_tokens
       maxValue: 64000
       minValue: 1
+provisioning: serverless
 sources:
     - https://platform.claude.com/docs/en/docs/about-claude/models
     - https://platform.claude.com/docs/en/about-claude/pricing

providers/aws-bedrock/eu.cohere.embed-v4:0.yaml

@@ -23,6 +23,7 @@ modalities:
         - embedding
 mode: embedding
 model: eu.cohere.embed-v4:0
+provisioning: serverless
 removeParams:
     - temperature
     - top_p

providers/aws-bedrock/eu.mistral.pixtral-large-2502-v1:0.yaml

@@ -14,6 +14,7 @@ costs:
 features:
     - function_calling
     - system_messages
+    - json_output
 limits:
     context_window: 128000
     max_input_tokens: 128000
@@ -33,11 +34,12 @@ params:
       key: max_tokens
       maxValue: 4096
       minValue: 1
+provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-mistral-pixtral-large.html
     - https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html
     - https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference-supported-models-features.html
-    - https://docs.mistral.ai/models/pixtral-large-24-11
+    - https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html
 status: active
 supportedModes:
     - chat

providers/aws-bedrock/global.amazon.nova-2-lite-v1:0.yaml

@@ -123,6 +123,7 @@ params:
       key: max_tokens
       maxValue: 65536
       minValue: 1
+provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/nova/latest/nova2-userguide/what-is-nova-2.html
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html

providers/aws-bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0.yaml

@@ -251,6 +251,7 @@ params:
       maxValue: 128000
       minValue: 1024
       type: number
+provisioning: serverless
 removeParams:
     - top_p
 sources:

providers/aws-bedrock/global.anthropic.claude-sonnet-4-20250514-v1:0.yaml

@@ -32,13 +32,15 @@ costs:
       output_cost_per_token: 0.000015
       output_cost_per_token_batches: 0.0000075
       region: ap-northeast-1
+deprecationDate: "2026-06-15"
 features:
     - function_calling
     - prompt_caching
     - cache_control
     - system_messages
     - tool_choice
     - assistant_prefill
+isDeprecated: true
 limits:
     context_window: 200000
     max_input_tokens: 200000

providers/aws-bedrock/global.anthropic.claude-sonnet-4-6.yaml

@@ -250,9 +250,11 @@ params:
       key: max_tokens
       maxValue: 64000
       minValue: 1
+provisioning: serverless
 sources:
     - https://platform.claude.com/docs/en/docs/about-claude/models
     - https://platform.claude.com/docs/en/about-claude/pricing
+    - https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference-support.html
 status: active
 supportedModes:
     - chat

providers/aws-bedrock/global.cohere.embed-v4:0.yaml

@@ -57,6 +57,7 @@ modalities:
         - embedding
 mode: embedding
 model: global.cohere.embed-v4:0
+provisioning: serverless
 removeParams:
     - max_tokens
     - temperature

providers/aws-bedrock/google.gemma-3-12b-it.yaml

@@ -29,6 +29,9 @@ costs:
     - input_cost_per_token: 1.1e-7
       output_cost_per_token: 3.5e-7
       region: ap-northeast-1
+features:
+    - function_calling
+    - system_messages
 limits:
     context_window: 128000
     max_input_tokens: 128000
@@ -47,8 +50,10 @@ params:
       key: max_tokens
       maxValue: 8192
       minValue: 1
+provisioning: serverless
 sources:
     - https://huggingface.co/google/gemma-3-12b-it
+    - https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-google-gemma-3-12b-it.html
 status: active
 supportedModes:
     - chat

providers/aws-bedrock/google.gemma-3-27b-it.yaml

@@ -49,6 +49,7 @@ params:
       key: max_tokens
       maxValue: 8192
       minValue: 1
+provisioning: serverless
 sources:
     - https://ai.google.dev/gemma/docs/core
     - https://huggingface.co/google/gemma-3-27b-it