update gpt-5 models

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@missionsquad/rosetta-ai",
-  "version": "1.11.4",
+  "version": "1.12.0",
   "description": "Unified TypeScript SDK for interacting with multiple AI providers (Anthropic, Google, Groq, OpenAI).",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",

src/core/mapping/common.utils.ts

@@ -83,6 +83,8 @@ export function mapBaseParams(
   temperature?: number
   topP?: number
   maxTokens?: number
+  reasoningEffort?: GenerateParams['reasoningEffort']
+  verbosity?: GenerateParams['verbosity']
   stopSequences?: string[]
 } {
   const stopSequences = Array.isArray(params.stop) ? params.stop : params.stop ? [params.stop] : undefined
@@ -102,12 +104,16 @@ export function mapBaseParams(
     temperature?: number
     topP?: number
     maxTokens?: number
+    reasoningEffort?: GenerateParams['reasoningEffort']
+    verbosity?: GenerateParams['verbosity']
     stopSequences?: string[]
   } = {}
 
   if (params.temperature !== undefined) result.temperature = params.temperature
   if (params.topP !== undefined) result.topP = params.topP
   if (effectiveMaxTokens !== undefined) result.maxTokens = effectiveMaxTokens
+  if (params.reasoningEffort !== undefined) result.reasoningEffort = params.reasoningEffort
+  if (params.verbosity !== undefined) result.verbosity = params.verbosity
   if (stopSequences !== undefined) result.stopSequences = stopSequences
 
   return result

src/core/mapping/gpt5.support.ts

@@ -0,0 +1,164 @@
+import type { ReasoningEffort } from '../../types'
+
+export interface Gpt5Support {
+  chatCompletionsSupported: boolean
+  allowedReasoningEfforts: ReasoningEffort[]
+  defaultReasoningEffort?: ReasoningEffort
+  fixedReasoningEffort?: ReasoningEffort
+  supportsVerbosity: boolean
+  supportsSampling: 'never' | 'always' | 'only_with_reasoning_none'
+}
+
+// Chat Completions-supported GPT-5 models verified for this implementation.
+const GPT5_CHAT_COMPLETIONS_SUPPORT: Record<string, Gpt5Support> = {
+  'gpt-5': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['minimal', 'low', 'medium', 'high'],
+    defaultReasoningEffort: 'medium',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5-mini': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['minimal', 'low', 'medium', 'high'],
+    defaultReasoningEffort: 'medium',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5-nano': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['minimal', 'low', 'medium', 'high'],
+    defaultReasoningEffort: 'medium',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5.1': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['none', 'low', 'medium', 'high'],
+    defaultReasoningEffort: 'none',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5.2': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['none', 'low', 'medium', 'high', 'xhigh'],
+    defaultReasoningEffort: 'none',
+    supportsVerbosity: true,
+    supportsSampling: 'only_with_reasoning_none'
+  },
+  'gpt-5.4': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['none', 'low', 'medium', 'high', 'xhigh'],
+    defaultReasoningEffort: 'none',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5-chat-latest': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: true,
+    supportsSampling: 'always'
+  },
+  'gpt-5.1-chat-latest': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['medium'],
+    fixedReasoningEffort: 'medium',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5.2-chat-latest': {
+    chatCompletionsSupported: true,
+    allowedReasoningEfforts: ['medium'],
+    fixedReasoningEffort: 'medium',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  }
+}
+
+// Verified GPT-5 inventory that is treated as Responses-only / unsupported on Chat Completions here.
+const GPT5_CHAT_COMPLETIONS_UNSUPPORTED: Record<string, Gpt5Support> = {
+  'gpt-5-codex': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5-pro': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: ['high'],
+    fixedReasoningEffort: 'high',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5-search-api': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.1-codex': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.1-codex-max': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.1-codex-mini': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.2-codex': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.2-pro': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: ['high'],
+    fixedReasoningEffort: 'high',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  },
+  'gpt-5.3-chat-latest': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.3-codex': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: [],
+    supportsVerbosity: false,
+    supportsSampling: 'never'
+  },
+  'gpt-5.4-pro': {
+    chatCompletionsSupported: false,
+    allowedReasoningEfforts: ['high'],
+    fixedReasoningEffort: 'high',
+    supportsVerbosity: true,
+    supportsSampling: 'never'
+  }
+}
+
+export const GPT5_SUPPORT_TABLE: Record<string, Gpt5Support> = {
+  ...GPT5_CHAT_COMPLETIONS_SUPPORT,
+  ...GPT5_CHAT_COMPLETIONS_UNSUPPORTED
+}
+
+const GPT5_SUPPORT_KEYS = Object.keys(GPT5_SUPPORT_TABLE).sort((a, b) => b.length - a.length)
+
+export function getGpt5Support(model: string): Gpt5Support | undefined {
+  const exact = GPT5_SUPPORT_TABLE[model]
+  if (exact) return exact
+
+  const prefix = GPT5_SUPPORT_KEYS.find(key => model.startsWith(`${key}-`))
+  return prefix ? GPT5_SUPPORT_TABLE[prefix] : undefined
+}

src/core/mapping/openai.mapper.ts

@@ -36,6 +36,7 @@ import { IProviderMapper } from './base.mapper'
 import { mapBaseParams, mapBaseToolChoice, mapToOpenAIResponseFormat } from './common.utils'
 import * as OpenAIEmbedMapper from './openai.embed.mapper'
 import * as OpenAIAudioMapper from './openai.audio.mapper'
+import { getGpt5Support } from './gpt5.support'
 import {
   isGPT5Model,
   isThinkingModel,
@@ -209,9 +210,65 @@ export class OpenAIMapper implements IProviderMapper {
     } else {
       delete basePayload.max_tokens
     }
+
     if (isGPT5) {
-      if (basePayload.temperature != null && !isNaN(basePayload.temperature)) {
-        basePayload.temperature = 1
+      const support = getGpt5Support(params.model!)
+
+      if (support?.chatCompletionsSupported === false) {
+        throw new UnsupportedFeatureError(this.provider, `Chat Completions for model '${params.model!}'`)
+      }
+
+      if (support) {
+        const requestedEffort = baseMappedParams.reasoningEffort
+        let effectiveEffort: GenerateParams['reasoningEffort'] | undefined
+
+        if (support.fixedReasoningEffort) {
+          effectiveEffort = support.fixedReasoningEffort
+        } else if (support.allowedReasoningEfforts.length > 0) {
+          if (
+            requestedEffort !== undefined &&
+            support.allowedReasoningEfforts.includes(requestedEffort)
+          ) {
+            effectiveEffort = requestedEffort
+          } else {
+            effectiveEffort = support.defaultReasoningEffort
+          }
+        }
+
+        if (effectiveEffort !== undefined) {
+          basePayload.reasoning_effort = effectiveEffort
+        } else {
+          delete basePayload.reasoning_effort
+        }
+
+        if (
+          support.supportsVerbosity &&
+          baseMappedParams.verbosity !== undefined &&
+          ['low', 'medium', 'high'].includes(baseMappedParams.verbosity)
+        ) {
+          basePayload.verbosity = baseMappedParams.verbosity
+        } else {
+          delete basePayload.verbosity
+        }
+
+        const allowsSampling =
+          support.supportsSampling === 'always' ||
+          (support.supportsSampling === 'only_with_reasoning_none' && effectiveEffort === 'none')
+
+        if (!allowsSampling) {
+          delete basePayload.temperature
+          delete basePayload.top_p
+        }
+
+        delete basePayload.max_tokens
+      } else {
+        // Conservative fallback for unknown future GPT-5-family models on Chat Completions:
+        // preserve token limit behavior but drop sampling and GPT-5-specific fields until modeled.
+        delete basePayload.temperature
+        delete basePayload.top_p
+        delete basePayload.reasoning_effort
+        delete basePayload.verbosity
+        delete basePayload.max_tokens
       }
     }
 

src/core/rosetta-ai.ts

@@ -727,6 +727,10 @@ export class RosettaAI {
       // --- Map Parameters ---
       const mappedParams = mapper.mapToProviderParams ? mapper.mapToProviderParams(effectiveParams) : effectiveParams
 
+      if (process.env.DEBUG === 'true' && providerKey === Provider.Anthropic) {
+          console.log(`[ROSETTA DEBUG] Mapped Anthropic stream request:\n${JSON.stringify(mappedParams, null, 2)}`)
+      }
+
       // --- Execute ---
       if (isCustom && mapper.executeStream && customConfig) {
         // Custom Provider Execution Path

src/types/params.types.ts

@@ -1,6 +1,9 @@
 import { ProviderKey, RosettaMessage, RosettaTool, RosettaAudioData, GenerateParamsProviderState } from './common.types'
 import { ProviderOptions } from './config.types'
 
+export type ReasoningEffort = 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh'
+export type Verbosity = 'low' | 'medium' | 'high'
+
 export type RosettaResponseFormat =
   | { type: 'text' }
   | {
@@ -44,6 +47,10 @@ export interface GenerateParams {
   temperature?: number
   /** Nucleus sampling parameter: considers only tokens comprising the top `topP` probability mass. */
   topP?: number
+  /** GPT-5 reasoning effort control for supported OpenAI Chat Completions models. */
+  reasoningEffort?: ReasoningEffort
+  /** GPT-5 verbosity control for supported OpenAI Chat Completions models. */
+  verbosity?: Verbosity
   /** Sequence(s) where the API will stop generating further tokens. */
   stop?: string | string[] | null
   /**