Sciorex Docs

English

Español

English

Español

Appearance

Models Reference ​

Sciorex supports multiple AI providers, giving you flexibility to choose the best model for your use case.

Supported Providers ​

ProviderCLI / TypeBest For
Claude CodeCLI (Cloud)Complex reasoning, coding, analysis
Google GeminiCLI (Cloud)Fast responses, multimodal, large context
OpenAI CodexCLI (Cloud)Code generation, general tasks
OpenCodeCLI (Cloud/Local)Open-source alternative CLI
LM StudioHTTP (Local)Privacy, offline, experimentation
OllamaHTTP (Local)Privacy, custom models, self-hosted

Setup Instructions

See Settings → AI Providers in the app for setup instructions for each provider.

Claude Models ​

Claude models excel at complex reasoning, coding tasks, and nuanced understanding.

ModelIDCapabilitiesContext
Claude Opus 4.6claude-opus-4-6Reasoning, Vision, Tools, PDF, Code200K
Claude Sonnet 5.0claude-sonnet-5-0Fast, Reasoning, Vision, Tools, PDF, Code200K
Claude Opus 4.5claude-opus-4-5-20251101Reasoning, Vision, Tools, PDF, Code200K
Claude Sonnet 4.5claude-sonnet-4-5-20250929Fast, Reasoning, Vision, Tools, PDF, Code200K
Claude Haiku 4.5claude-haiku-4-5-20251001Fast, Vision, Tools, Code200K
Claude Opus 4.1claude-opus-4-1-20250805Reasoning, Vision, Tools, PDF, Code200K
Claude Opus 4claude-opus-4-20250514Reasoning, Vision, Tools, PDF, Code200K
Claude Sonnet 4claude-sonnet-4-20250514Reasoning, Vision, Tools, PDF, Code200K
Claude Haiku 3.5claude-3-5-haiku-20241022Fast, Vision, Tools, Code200K

Default Choice

Claude Sonnet 5.0 is the recommended default for most coding tasks.

Gemini Models ​

Gemini models offer fast responses and excellent multimodal capabilities.

ModelIDCapabilitiesContext
Gemini 3 Pro Previewgemini-3-pro-previewReasoning, Vision, Tools, PDF, Code1M
Gemini 3 Flash Previewgemini-3-flash-previewFast, Vision, Tools, PDF, Code1M
Gemini 2.5 Progemini-2.5-proReasoning, Vision, Tools, PDF, Code1M
Gemini 2.5 Flashgemini-2.5-flashFast, Vision, Tools, PDF, Code1M
Gemini 2.5 Flash Litegemini-2.5-flash-liteFast, Tools, Code128K

Large Context

Gemini models support up to 1M tokens context window.

OpenAI Codex Models ​

OpenAI Codex models provide strong coding capabilities with large context windows.

Codex Models ​

ModelIDCapabilitiesContext
GPT-5.3 Codexgpt-5.3-codexReasoning, Tools, PDF, Code400K
GPT-5.2 Codexgpt-5.2-codexReasoning, Tools, PDF, Code400K
GPT-5.1 Codex Maxgpt-5.1-codex-maxReasoning, Tools, PDF, Code400K
GPT-5.1 Codexgpt-5.1-codexReasoning, Tools, PDF, Code400K
GPT-5.1 Codex Minigpt-5.1-codex-miniFast, Tools, Code400K
GPT-5 Codexgpt-5-codexReasoning, Tools, PDF, Code400K
GPT-5 Codex Minigpt-5-codex-miniFast, Tools, Code400K

GPT Models ​

ModelIDCapabilitiesContext
GPT-5.2gpt-5.2Reasoning, Vision, Tools, PDF, Code400K
GPT-5.1gpt-5.1Reasoning, Vision, Tools, PDF, Code400K
GPT-5gpt-5Reasoning, Vision, Tools, PDF, Code400K

GPT-OSS Models (Local) ​

These models run locally via Codex CLI with the --oss flag.

ModelIDCapabilitiesContext
GPT-OSS 120Bopenai/gpt-oss-120bReasoning, Code, Tools128K
GPT-OSS 20Bopenai/gpt-oss-20bFast, Code, Tools128K

OpenCode ​

OpenCode is an open-source CLI that provides an alternative interface to various AI providers. Models are configured through OpenCode's own settings.

TIP

OpenCode is useful for developers who prefer an open-source CLI tool or need to connect to alternative API endpoints.

Local Models (LM Studio & Ollama) ​

Run models locally for privacy, offline work, or cost savings. Models are fetched dynamically from your local server.

LM Studio ​

Popular models for LM Studio:

ModelParametersBest ForVRAM Required
Qwen 2.5 Coder7B / 32BCode generation8GB / 24GB
DeepSeek Coder V216BAdvanced coding12GB
CodeLlama7B / 34BCode completion8GB / 24GB
Llama 3.23B / 11BGeneral tasks4GB / 12GB

Ollama ​

Popular models for Ollama:

ModelBest For
qwen2.5-coderCode generation
deepseek-coder-v2Advanced coding
codellamaCode tasks
llama3.2General tasks
mistralFast responses

Hardware Requirements

  • 7B models: 8GB+ VRAM
  • 13B models: 16GB+ VRAM
  • 34B+ models: 24GB+ VRAM

Model Capabilities ​

CapabilityDescription
FastOptimized for speed
VisionCan analyze images
ReasoningAdvanced reasoning capabilities
ToolsCan use tools and function calling
PDFCan read and analyze PDFs
CodeOptimized for code generation
Long ContextLarge context window
Image GenCan generate images

Provider Comparison ​

By Use Case ​

Use CaseRecommended ProviderRecommended Model
Complex architectureClaudeOpus 4.6
General codingClaude / GeminiSonnet 5.0 / Gemini 3 Flash
Quick tasksGemini / CodexGemini 2.5 Flash Lite / Codex Mini
Code reviewClaudeSonnet 5.0
Privacy-sensitiveLocalLM Studio / Ollama
Offline workLocalAny local model
Long contextGemini / CodexGemini 3 Pro (1M) / GPT-5.3 (400K)

By Feature ​

FeatureClaudeGeminiCodexOpenCodeLocal
Extended ThinkingYesYesYesVariesNo
Effort LevelsYesNoNoNoNo
Tool UseYesYesYesYesVaries
Vision/ImagesYesYesYesVariesVaries
Max Context200K1M400KVariesVaries
OfflineNoNoNoNoYes

Extended Thinking ​

Extended Thinking allows models to reason through complex problems before responding.

LevelToken BudgetClaudeGeminiCodex
Off0YesYesYes
Think1,024YesYesYes
Think Hard10,000YesYesYes
Think Harder16,000Yes--
Ultrathink32,000Yes--

Effort Levels ​

Claude Opus 4.6 supports Effort Levels for fine-tuning reasoning depth:

LevelDescription
LowMinimal reasoning, fastest responses
MediumBalanced reasoning
HighThorough analysis
MaxDeepest reasoning available

Per-Provider Reasoning Systems ​

Each provider has its own reasoning controls beyond Extended Thinking:

Claude (Effort Levels):

LevelValueDescription
LowlowMinimal reasoning
MediummediumBalanced
HighhighThorough (default)
MaxmaxMaximum depth (Opus 4.6 only)

Codex (Reasoning Effort):

LevelValueDescription
NonenoneNo reasoning
MinimalminimalLightest reasoning
LowlowBasic reasoning
MediummediumBalanced
HighhighDeep reasoning
Extra HighxhighMaximum reasoning

OpenCode (Variants):

VariantValueDescription
OffoffStandard mode
HighhighEnhanced reasoning
MaxmaxMaximum reasoning

Unified Thinking Levels ​

Sciorex normalizes thinking levels across providers for features like Council Mode:

Unified LevelClaudeGeminiCodex
offOffOffNone
lightThinkThinkLow
mediumThink HardThink HardMedium
hardThink Harder-High
maxUltrathink-Extra High

Customizable Budgets

You can customize thinking token budgets in Settings → AI Providers → Thinking Budgets.

Configuring in Agents ​

yaml
name: Code Reviewer
adapter: claude-code
model: claude-sonnet-5-0
thinkingLevel: think
yaml
name: Fast Analyzer
adapter: google-gemini
model: gemini-3-flash-preview
yaml
name: Local Formatter
adapter: ollama
model: qwen2.5-coder

Pricing ​

INFO

Pricing is set by each AI provider and may change. Check each provider's website for current pricing.

Claude ​

ModelInputOutput
Opus 4.6$15/1M$75/1M
Sonnet 5.0$3/1M$15/1M
Opus 4.5$15/1M$75/1M
Sonnet 4.5$3/1M$15/1M
Haiku 4.5$1/1M$5/1M
Haiku 3.5$0.80/1M$4/1M

Gemini ​

ModelInputOutput
Gemini 3 Pro$1.25/1M$5/1M
Gemini 3 Flash$0.075/1M$0.30/1M
Gemini 2.5 Pro$1.25/1M$5/1M
Gemini 2.5 Flash$0.075/1M$0.30/1M
Gemini 2.5 Flash Lite$0.019/1M$0.075/1M

Codex ​

ModelInputOutput
GPT-5.3 Codex$1.25/1M$10/1M
GPT-5.2 Codex$1.25/1M$10/1M
GPT-5.2$1.75/1M$14/1M
GPT-5.1 Codex Max$1.25/1M$10/1M
GPT-5.1 Codex Mini$0.25/1M$2/1M

Best Practices ​

For Agents ​

yaml
# Complex analysis
name: Senior Reviewer
adapter: claude-code
model: claude-opus-4-6
thinkingLevel: think-hard

# Fast local tasks
name: Code Formatter
adapter: ollama
model: qwen2.5-coder

# Quick cloud analysis
name: Quick Analyzer
adapter: google-gemini
model: gemini-3-flash-preview

For Flows ​

  • Use local models or Gemini Flash for transformation nodes
  • Use Claude Sonnet or GPT-5.3 for decision nodes
  • Use Claude Opus for critical analysis nodes

Sciorex is proprietary software.

Copyright © 2025-2026 Sciorex