Reasoning effort fix (#1674)

harshiv-26 · web-flow · commit 1f67f61b5b0a · 2026-07-02T16:46:21.000+05:30
* mantle: reasoning-effort fix

* fix for vertex and cohere

* fix: strip reasoning_effort on vertex claude-sonnet-4-5@20250929 to match sonnet-4-5

* cohere
diff --git a/providers/aws-bedrock-mantle/openai.gpt-oss-20b.yaml b/providers/aws-bedrock-mantle/openai.gpt-oss-20b.yaml
@@ -53,6 +53,13 @@ modalities:
         - text
 mode: chat
 model: openai.gpt-oss-20b
+params:
+    - key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+      type: string
 provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-20b.html
diff --git a/providers/aws-bedrock-mantle/openai.gpt-oss-safeguard-120b.yaml b/providers/aws-bedrock-mantle/openai.gpt-oss-safeguard-120b.yaml
@@ -51,6 +51,13 @@ modalities:
         - text
 mode: chat
 model: openai.gpt-oss-safeguard-120b
+params:
+    - key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+      type: string
 provisioning: serverless
 sources:
     - https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-safeguard-120b.html
diff --git a/providers/cohere/cohere-transcribe-03-2026.yaml b/providers/cohere/cohere-transcribe-03-2026.yaml
@@ -10,6 +10,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/transcribe
 status: active
diff --git a/providers/cohere/default.yaml b/providers/cohere/default.yaml
@@ -20,3 +20,9 @@ params:
     - defaultValue: true
       key: stream
       type: boolean
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - none
+          - high
+      type: string
diff --git a/providers/cohere/embed-english-light-v3.0.yaml b/providers/cohere/embed-english-light-v3.0.yaml
@@ -19,6 +19,7 @@ removeParams:
     - temperature
     - stream
     - max_tokens
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/reference/embed
diff --git a/providers/cohere/embed-english-v3.0-image.yaml b/providers/cohere/embed-english-v3.0-image.yaml
@@ -13,6 +13,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/reference/embed
diff --git a/providers/cohere/embed-english-v3.0.yaml b/providers/cohere/embed-english-v3.0.yaml
@@ -20,6 +20,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/reference/embed
     - https://docs.cohere.com/docs/cohere-embed
diff --git a/providers/cohere/embed-multilingual-light-v3.0-image.yaml b/providers/cohere/embed-multilingual-light-v3.0-image.yaml
@@ -14,6 +14,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/reference/embed
diff --git a/providers/cohere/embed-multilingual-light-v3.0.yaml b/providers/cohere/embed-multilingual-light-v3.0.yaml
@@ -19,6 +19,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/v2/reference/embed
diff --git a/providers/cohere/embed-multilingual-v3.0-image.yaml b/providers/cohere/embed-multilingual-v3.0-image.yaml
@@ -15,6 +15,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/reference/embed
     - https://docs.cohere.com/docs/cohere-embed
diff --git a/providers/cohere/embed-multilingual-v3.0.yaml b/providers/cohere/embed-multilingual-v3.0.yaml
@@ -19,6 +19,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/reference/embed
diff --git a/providers/cohere/embed-v4.0.yaml b/providers/cohere/embed-v4.0.yaml
@@ -20,6 +20,7 @@ removeParams:
     - temperature
     - stream
     - max_tokens
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/cohere-embed
     - https://docs.cohere.com/reference/embed
diff --git a/providers/cohere/rerank-english-v3.0.yaml b/providers/cohere/rerank-english-v3.0.yaml
@@ -15,6 +15,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/rerank
     - https://docs.cohere.com/reference/rerank
diff --git a/providers/cohere/rerank-multilingual-v3.0.yaml b/providers/cohere/rerank-multilingual-v3.0.yaml
@@ -15,6 +15,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/rerank
     - https://docs.cohere.com/reference/rerank
diff --git a/providers/cohere/rerank-v3.5.yaml b/providers/cohere/rerank-v3.5.yaml
@@ -16,6 +16,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/rerank
     - https://docs.cohere.com/v2/reference/rerank
diff --git a/providers/cohere/rerank-v4.0-fast.yaml b/providers/cohere/rerank-v4.0-fast.yaml
@@ -10,6 +10,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/docs/rerank
     - https://docs.cohere.com/reference/rerank
diff --git a/providers/cohere/rerank-v4.0-pro.yaml b/providers/cohere/rerank-v4.0-pro.yaml
@@ -15,6 +15,7 @@ removeParams:
     - max_tokens
     - temperature
     - stream
+    - reasoning_effort
 sources:
     - https://docs.cohere.com/reference/rerank
     - https://docs.cohere.com/docs/rerank
diff --git a/providers/google-vertex/anthropic/claude-fable-5.yaml b/providers/google-vertex/anthropic/claude-fable-5.yaml
@@ -44,6 +44,15 @@ params:
       key: max_tokens
       maxValue: 128000
       minValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - temperature
diff --git a/providers/google-vertex/anthropic/claude-haiku-4-5.yaml b/providers/google-vertex/anthropic/claude-haiku-4-5.yaml
@@ -66,6 +66,7 @@ params:
 provisioning: serverless
 removeParams:
     - "n"
+    - reasoning_effort
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/haiku-4-5
     - https://platform.claude.com/docs/en/about-claude/models
diff --git a/providers/google-vertex/anthropic/claude-haiku-4-5@20251001.yaml b/providers/google-vertex/anthropic/claude-haiku-4-5@20251001.yaml
@@ -55,6 +55,8 @@ params:
     - key: max_tokens
       maxValue: 64000
 provisioning: serverless
+removeParams:
+    - reasoning_effort
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/haiku-4-5
     - https://platform.claude.com/docs/en/about-claude/pricing
diff --git a/providers/google-vertex/anthropic/claude-opus-4-1.yaml b/providers/google-vertex/anthropic/claude-opus-4-1.yaml
@@ -42,6 +42,15 @@ model: anthropic/claude-opus-4-1
 params:
     - key: max_tokens
       maxValue: 32000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 retirementDate: "2026-08-05"
 sources:
diff --git a/providers/google-vertex/anthropic/claude-opus-4-1@20250805.yaml b/providers/google-vertex/anthropic/claude-opus-4-1@20250805.yaml
@@ -40,6 +40,15 @@ model: anthropic/claude-opus-4-1@20250805
 params:
     - key: max_tokens
       maxValue: 32000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 retirementDate: "2026-08-05"
 sources:
diff --git a/providers/google-vertex/anthropic/claude-opus-4-5.yaml b/providers/google-vertex/anthropic/claude-opus-4-5.yaml
@@ -57,6 +57,13 @@ model: anthropic/claude-opus-4-5
 params:
     - key: max_tokens
       maxValue: 64000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+      type: string
 provisioning: serverless
 retirementDate: "2026-11-24"
 sources:
diff --git a/providers/google-vertex/anthropic/claude-opus-4-5@20251101.yaml b/providers/google-vertex/anthropic/claude-opus-4-5@20251101.yaml
@@ -62,6 +62,13 @@ model: anthropic/claude-opus-4-5@20251101
 params:
     - key: max_tokens
       maxValue: 64000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+      type: string
 provisioning: serverless
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/opus-4-5
diff --git a/providers/google-vertex/anthropic/claude-opus-4-6.yaml b/providers/google-vertex/anthropic/claude-opus-4-6.yaml
@@ -65,6 +65,14 @@ params:
       key: max_tokens
       maxValue: 128000
       minValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - max
+      type: string
 provisioning: serverless
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/opus-4-6
diff --git a/providers/google-vertex/anthropic/claude-opus-4-6@default.yaml b/providers/google-vertex/anthropic/claude-opus-4-6@default.yaml
@@ -77,6 +77,14 @@ params:
       maxValue: 128000
     - key: temperature
       maxValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - "n"
diff --git a/providers/google-vertex/anthropic/claude-opus-4-7.yaml b/providers/google-vertex/anthropic/claude-opus-4-7.yaml
@@ -49,6 +49,15 @@ model: anthropic/claude-opus-4-7
 params:
     - key: max_tokens
       maxValue: 128000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - "n"
diff --git a/providers/google-vertex/anthropic/claude-opus-4-8.yaml b/providers/google-vertex/anthropic/claude-opus-4-8.yaml
@@ -44,6 +44,15 @@ params:
       key: max_tokens
       maxValue: 128000
       minValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - temperature
diff --git a/providers/google-vertex/anthropic/claude-opus-4.yaml b/providers/google-vertex/anthropic/claude-opus-4.yaml
@@ -43,6 +43,15 @@ params:
       key: max_tokens
       maxValue: 32000
       minValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/opus-4
diff --git a/providers/google-vertex/anthropic/claude-opus-4@20250514.yaml b/providers/google-vertex/anthropic/claude-opus-4@20250514.yaml
@@ -39,6 +39,15 @@ model: anthropic/claude-opus-4@20250514
 params:
     - key: max_tokens
       maxValue: 32000
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 sources:
     - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4-5.yaml b/providers/google-vertex/anthropic/claude-sonnet-4-5.yaml
@@ -149,6 +149,8 @@ params:
       maxValue: 64000
       minValue: 1
 provisioning: serverless
+removeParams:
+    - reasoning_effort
 sources:
     - https://platform.claude.com/docs/en/about-claude/models/overview
     - https://platform.claude.com/docs/en/about-claude/pricing
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4-5@20250929.yaml b/providers/google-vertex/anthropic/claude-sonnet-4-5@20250929.yaml
@@ -148,6 +148,8 @@ params:
     - key: max_tokens
       maxValue: 64000
 provisioning: serverless
+removeParams:
+    - reasoning_effort
 sources:
     - https://platform.claude.com/docs/en/about-claude/models
     - https://platform.claude.com/docs/en/about-claude/pricing
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4-6.yaml b/providers/google-vertex/anthropic/claude-sonnet-4-6.yaml
@@ -60,6 +60,15 @@ params:
       maxValue: 64000
     - key: temperature
       maxValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - "n"
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4-6@default.yaml b/providers/google-vertex/anthropic/claude-sonnet-4-6@default.yaml
@@ -76,6 +76,15 @@ params:
       maxValue: 64000
     - key: temperature
       maxValue: 1
+    - defaultValue: high
+      key: reasoning_effort
+      supportedValues:
+          - low
+          - medium
+          - high
+          - xhigh
+          - max
+      type: string
 provisioning: serverless
 removeParams:
     - "n"
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4.yaml b/providers/google-vertex/anthropic/claude-sonnet-4.yaml
diff --git a/providers/google-vertex/anthropic/claude-sonnet-4@20250514.yaml b/providers/google-vertex/anthropic/claude-sonnet-4@20250514.yaml
