github/sim
1
1
Fork 0
mirror of https://github.com/simstudioai/sim.git synced 2026-04-06 03:00:16 -04:00
Code Issues Packages Projects Releases Wiki Activity

improvement(providers): audit and update all provider model definitions (#3893)

Browse Source 
* improvement(providers): audit and update all provider model definitions

* fix(providers): add maxOutputTokens to azure/o3 and azure/o4-mini

* fix(providers): move maxOutputTokens inside capabilities for azure models
This commit is contained in:
Waleed
2026-04-01 19:32:16 -07:00
committed by GitHub
parent 27a11a269d
commit 4c94f3cf78
3 changed files with 414 additions and 305 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
apps/sim/providers/anthropic/core.ts
View File

@@ -77,17 +77,24 @@ const THINKING_BUDGET_TOKENS: Record<string, number> = {
}
/**
* Checks if a model supports adaptive thinking (Opus 4.6+)
* Checks if a model supports adaptive thinking (thinking.type: "adaptive").
* Per the Anthropic API, only Opus 4.6 and Sonnet 4.6 support adaptive thinking.
* Opus 4.5 supports effort but NOT adaptive thinking — it uses budget_tokens with type: "enabled".
*/
function supportsAdaptiveThinking(modelId: string): boolean {
const normalizedModel = modelId.toLowerCase()
return normalizedModel.includes('opus-4-6') || normalizedModel.includes('opus-4.6')
return (
normalizedModel.includes('opus-4-6') ||
normalizedModel.includes('opus-4.6') ||
normalizedModel.includes('sonnet-4-6') ||
normalizedModel.includes('sonnet-4.6')
)
}
/**
* Builds the thinking configuration for the Anthropic API based on model capabilities and level.
*
* - Opus 4.6: Uses adaptive thinking with effort parameter (recommended by Anthropic)
* - Opus 4.6, Sonnet 4.6: Uses adaptive thinking with effort parameter
* - Other models: Uses budget_tokens-based extended thinking
*
* Returns both the thinking config and optional output_config for adaptive thinking.
@@ -104,7 +111,7 @@ function buildThinkingConfig(
return null
}
// Opus 4.6 uses adaptive thinking with effort parameter
// Models with effort support use adaptive thinking
if (supportsAdaptiveThinking(modelId)) {
return {
thinking: { type: 'adaptive' },

682
apps/sim/providers/models.ts
View File

File diff suppressed because it is too large Load Diff

22
apps/sim/providers/utils.test.ts
View File

@@ -392,7 +392,6 @@ describe('Model Capabilities', () => {
expect(supportsThinking('claude-sonnet-4-5')).toBe(true)
expect(supportsThinking('claude-sonnet-4-0')).toBe(true)
expect(supportsThinking('claude-haiku-4-5')).toBe(true)
expect(supportsThinking('gemini-3-pro-preview')).toBe(true)
expect(supportsThinking('gemini-3-flash-preview')).toBe(true)
})
@@ -511,7 +510,6 @@ describe('Model Capabilities', () => {
expect(MODELS_WITH_THINKING).toContain('claude-sonnet-4-5')
expect(MODELS_WITH_THINKING).toContain('claude-sonnet-4-0')
expect(MODELS_WITH_THINKING).toContain('gemini-3-pro-preview')
expect(MODELS_WITH_THINKING).toContain('gemini-3-flash-preview')
expect(MODELS_WITH_THINKING).toContain('claude-haiku-4-5')
@@ -523,7 +521,12 @@ describe('Model Capabilities', () => {
it.concurrent('should have GPT-5 models in both reasoning effort and verbosity arrays', () => {
const gpt5ModelsWithReasoningEffort = MODELS_WITH_REASONING_EFFORT.filter(
(m) => m.includes('gpt-5') && !m.includes('chat-latest') && !m.includes('gpt-5.4-pro')
(m) =>
m.includes('gpt-5') &&
!m.includes('chat-latest') &&
!m.includes('gpt-5.4-pro') &&
!m.includes('gpt-5.2-pro') &&
!m.includes('gpt-5-pro')
)
const gpt5ModelsWithVerbosity = MODELS_WITH_VERBOSITY.filter(
(m) => m.includes('gpt-5') && !m.includes('chat-latest')
@@ -533,6 +536,12 @@ describe('Model Capabilities', () => {
expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5.4-pro')
expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5.4-pro')
expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5.2-pro')
expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5.2-pro')
expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5-pro')
expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5-pro')
expect(MODELS_WITH_REASONING_EFFORT).toContain('o1')
expect(MODELS_WITH_VERBOSITY).not.toContain('o1')
})
@@ -629,11 +638,6 @@ describe('Model Capabilities', () => {
})
it.concurrent('should return correct levels for Gemini 3 models', () => {
const proLevels = getThinkingLevelsForModel('gemini-3-pro-preview')
expect(proLevels).toBeDefined()
expect(proLevels).toContain('low')
expect(proLevels).toContain('high')
const flashLevels = getThinkingLevelsForModel('gemini-3-flash-preview')
expect(flashLevels).toBeDefined()
expect(flashLevels).toContain('minimal')
@@ -669,7 +673,7 @@ describe('Max Output Tokens', () => {
})
it.concurrent('should return correct max for Claude Opus 4.1', () => {
expect(getMaxOutputTokensForModel('claude-opus-4-1')).toBe(64000)
expect(getMaxOutputTokensForModel('claude-opus-4-1')).toBe(32000)
})
it.concurrent('should return standard default for models without maxOutputTokens', () => {