Use provider-aware token parameter for GPT-5-family chat completions

Use max_completion_tokens for OpenAI GPT-5-family Chat Completions and
keep max_tokens for non-openai or non-GPT-5-family paths.

Observed error on GPT-5.1-family responses:
Unsupported parameter: 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead.

Author: PeterDaveHello

src/services/apis/custom-api.mjs

@@ -10,6 +10,7 @@ import { fetchSSE } from '../../utils/fetch-sse.mjs'
 import { getConversationPairs } from '../../utils/get-conversation-pairs.mjs'
 import { isEmpty } from 'lodash-es'
 import { pushRecord, setAbortController } from './shared.mjs'
+import { getChatCompletionsTokenParams } from './openai-token-params.mjs'
 
 /**
  * @param {Browser.Runtime.Port} port
@@ -55,7 +56,7 @@ export async function generateAnswersWithCustomApi(
       messages: prompt,
       model: modelName,
       stream: true,
-      max_tokens: config.maxResponseTokenLength,
+      ...getChatCompletionsTokenParams('custom', modelName, config.maxResponseTokenLength),
       temperature: config.temperature,
     }),
     onMessage(message) {

src/services/apis/openai-api.mjs

@@ -6,6 +6,7 @@ import { getConversationPairs } from '../../utils/get-conversation-pairs.mjs'
 import { isEmpty } from 'lodash-es'
 import { getCompletionPromptBase, pushRecord, setAbortController } from './shared.mjs'
 import { getModelValue } from '../../utils/model-name-convert.mjs'
+import { getChatCompletionsTokenParams } from './openai-token-params.mjs'
 
 /**
  * @param {Browser.Runtime.Port} port
@@ -103,6 +104,8 @@ export async function generateAnswersWithChatgptApi(port, question, session, api
     question,
     session,
     apiKey,
+    {},
+    'openai',
   )
 }
 
@@ -113,6 +116,7 @@ export async function generateAnswersWithChatgptApiCompat(
   session,
   apiKey,
   extraBody = {},
+  provider = 'compat',
 ) {
   const { controller, messageListener, disconnectListener } = setAbortController(port)
   const model = getModelValue(session)
@@ -123,6 +127,12 @@ export async function generateAnswersWithChatgptApiCompat(
     false,
   )
   prompt.push({ role: 'user', content: question })
+  const tokenParams = getChatCompletionsTokenParams(provider, model, config.maxResponseTokenLength)
+  const conflictingTokenParamKey =
+    'max_completion_tokens' in tokenParams ? 'max_tokens' : 'max_completion_tokens'
+  // Avoid sending both token-limit fields when caller passes extraBody.
+  const safeExtraBody = { ...extraBody }
+  delete safeExtraBody[conflictingTokenParamKey]
 
   let answer = ''
   let finished = false
@@ -143,9 +153,9 @@ export async function generateAnswersWithChatgptApiCompat(
       messages: prompt,
       model,
       stream: true,
-      max_tokens: config.maxResponseTokenLength,
+      ...tokenParams,
       temperature: config.temperature,
-      ...extraBody,
+      ...safeExtraBody,
     }),
     onMessage(message) {
       console.debug('sse message', message)

src/services/apis/openai-token-params.mjs

@@ -0,0 +1,21 @@
+const GPT5_CHAT_COMPLETIONS_MODEL_PATTERN = /(^|\/)gpt-5([.-]|$)/
+
+function shouldUseMaxCompletionTokens(provider, model) {
+  const normalizedProvider = String(provider || '').toLowerCase()
+  const normalizedModel = String(model || '').toLowerCase()
+
+  switch (true) {
+    case normalizedProvider === 'openai' &&
+      GPT5_CHAT_COMPLETIONS_MODEL_PATTERN.test(normalizedModel):
+      return true
+    default:
+      return false
+  }
+}
+
+export function getChatCompletionsTokenParams(provider, model, maxResponseTokenLength) {
+  if (shouldUseMaxCompletionTokens(provider, model))
+    return { max_completion_tokens: maxResponseTokenLength }
+
+  return { max_tokens: maxResponseTokenLength }
+}

src/services/apis/openai-token-params.test.mjs

@@ -0,0 +1,57 @@
+import test from 'node:test'
+import assert from 'node:assert/strict'
+import { getChatCompletionsTokenParams } from './openai-token-params.mjs'
+
+test('uses max_completion_tokens for gpt-5.x chat models', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', 'gpt-5.2-chat-latest', 1024), {
+    max_completion_tokens: 1024,
+  })
+})
+
+test('uses max_completion_tokens for provider-prefixed gpt-5.x models', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', 'openai/gpt-5.2', 2048), {
+    max_completion_tokens: 2048,
+  })
+})
+
+test('uses max_completion_tokens for gpt-5 baseline model name', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', 'gpt-5', 1536), {
+    max_completion_tokens: 1536,
+  })
+})
+
+test('uses max_tokens for non gpt-5 chat models', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', 'gpt-4o', 512), {
+    max_tokens: 512,
+  })
+})
+
+test('uses max_tokens for lookalike model names', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', 'my-gpt-5-clone', 640), {
+    max_tokens: 640,
+  })
+})
+
+test('uses max_tokens for empty model values', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('openai', '', 256), {
+    max_tokens: 256,
+  })
+})
+
+test('uses max_tokens for non OpenAI providers even with gpt-5 models', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('some-proxy-provider', 'openai/gpt-5.2', 257), {
+    max_tokens: 257,
+  })
+})
+
+test('uses max_completion_tokens for mixed-case OpenAI provider and model', () => {
+  assert.deepEqual(getChatCompletionsTokenParams('OpenAI', 'GPT-5.1', 258), {
+    max_completion_tokens: 258,
+  })
+})
+
+test('uses max_tokens when provider is undefined', () => {
+  assert.deepEqual(getChatCompletionsTokenParams(undefined, 'gpt-5.1', 259), {
+    max_tokens: 259,
+  })
+})