Models, role routing, and agent skills.

21 live verified upstream models across Flagship Reasoning, Coding, Vision & Multimodal Chat, and Low-Latency Tool Execution, plus 7 cognitive role bindings and the tiered skills library.

Varynth public models

21 verified & active, no account needed — click any row to copy its model ID

  • varynth/gpt-oss-120bOpenAI GPT-OSS 120B (Reasoning · Tools)reasoning
  • varynth/qwen3.8-27bQwen 3.8 27B (Vision · Reasoning · Tools)vision
  • varynth/nemotron-3-120b-a12bNVIDIA Nemotron 3 120B A12B (Reasoning · Tools)reasoning
  • varynth/qwen3-30b-a3b-fp8Qwen 3 30B A3B FP8 (Reasoning · Tools)reasoning
  • varynth/qwen2.5-coder-32b-instructQwen 2.5 Coder 32B Instruct (Coding)coding
  • varynth/deepseek-r1-distill-qwen-32bDeepSeek R1 Distill Qwen 32B (Reasoning)reasoning
  • varynth/qwq-32bQwen QwQ 32B (Reasoning)reasoning
  • varynth/llama-4-scout-17b-16e-instructMeta Llama 4 Scout 17B 16E (Vision · Tools)vision
  • varynth/llama-3.3-70b-instruct-fp8-fastMeta Llama 3.3 70B Instruct FP8 Fast (Tools)coding
  • varynth/gemma-4-26b-a4b-itGoogle Gemma 4 26B A4B IT (Vision · Reasoning · Tools)vision
  • varynth/glm-4.7-flashZ.AI GLM 4.7 Flash (Reasoning · Tools)reasoning
  • varynth/mistral-small-3.1-24b-instructMistral Small 3.1 24B Instruct (Vision · Tools)vision
  • varynth/gpt-oss-20bOpenAI GPT-OSS 20B (Fast · Reasoning · Tools)reasoning
  • varynth/gemma-sea-lion-v4-27b-itAI Singapore Gemma SEA-LION v4 27B ITfast
  • varynth/llama-3.1-8b-instruct-fastMeta Llama 3.1 8B Instruct Fastfast
  • varynth/llama-3.1-8b-instruct-fp8Meta Llama 3.1 8B Instruct FP8fast
  • varynth/granite-4.0-h-microIBM Granite 4.0 H Micro (Fast · Tools)coding
  • varynth/llama-3.2-3b-instructMeta Llama 3.2 3B Instructfast
  • varynth/llama-3.2-1b-instructMeta Llama 3.2 1B Instructfast
  • varynth/gemma-7b-it-loraGoogle Gemma 7B IT LoRAfast
  • varynth/gemma-2b-it-loraGoogle Gemma 2B IT LoRAfast

Multi-role /model binding matrix

7 specialized slots in ~/.varynth/config.toml

1. Default & 2. Plan

default drives daily coding and multi-round tool execution (up to 128 rounds per turn). plan handles architectural synthesis when you invoke /plan or cycle into Plan mode with Shift+Tab.

/model : press 1 (default) or 2 (plan)

3. Smol & 4. Advisor

smol runs low-latency context compaction (/compact) and session titling. advisor acts as an independent verification auditor for /review and autonomous /goal loops.

/model : press 3 (smol) or 4 (advisor)

5. Widget, 6. Memory & 7. Fallback

widget and memory power UI artifacts and Obsidian .varynth/MEMORY/ indexing, while fallback automatically takes over if the primary provider returns an upstream error.

/model : press 5, 6, or 7 (fallback)

Tiered agent skills catalog

12 skills, active tier: free

Create Skill

.agents/skills/create-skill
free

Scaffold and develop new agent skills with standard YAML frontmatter, deterministic execution, and unit tests.

Entitled and activeready

CLI Design

.agents/skills/cli-design
free

Guidelines and ergonomic patterns for building professional, responsive terminal user interfaces and command-line tools.

Entitled and activeready

Backend Core

.agents/skills/backend-core
free

Core backend architecture patterns, asynchronous runtime management, database connection pooling, and resilient API design.

Entitled and activeready

UI Foundations

.agents/skills/ui-foundations
free

Foundational design system and styling standards for clean, high-performance web and desktop user interfaces.

Entitled and activeready

Computer Use

.agents/skills/computer-use
free

Codex-style visual overlay computer-use operator with DPI scaling, multi-monitor awareness, glowing aim halo, click ripple, live operator pill, and safety kill-switch.

Entitled and activeready

Motion Fream

.agents/skills/motion-fream
free

Deterministic frame-by-frame HTML/CSS/Canvas motion graphics & 60fps launch video generator (Playwright + FFmpeg).

Entitled and activeready

Create MCP

.agents/skills/create-mcp
pro

Scaffold and build Model Context Protocol (MCP) servers and tools for Varynth.

Requires Pro planlocked

Create Plugin

.agents/skills/create-plugin
pro

Design, package, and distribute native Varynth extensions and plugins.

Requires Pro planlocked

Fullstack Architect

.agents/skills/fullstack-architect
pro

End-to-end fullstack systems architecture, zero-downtime migrations, and distributed scalability.

Requires Pro planlocked

Generative UI Pro

.agents/skills/generative-ui-pro
pro

Next-gen generative UI with dynamic streaming component synthesis and reactive state binding.

Requires Pro planlocked

Motion Fream Pro

.agents/skills/motion-fream-pro
pro

Pro automated deterministic 60fps motion graphics & launch video studio with Flux AI hero asset generation and Playwright + FFmpeg rendering.

Requires Pro planlocked

Security Auditor

.agents/skills/security-auditor
pro

Advanced static analysis, secret scanning, dependency supply chain audit, and memory safety verification.

Requires Pro planlocked

Provider-grouped account catalog

sign in to load, 9 active provider families

Sign in to browse models grouped by provider family.

Signed-in accounts can call any verified model by its provider-prefixed identifier (openai/gpt-oss-120b, qwen/qwen3.8-27b, nvidia/nemotron-3-120b-a12b, deepseek/deepseek-r1-distill-qwen-32b, meta/llama-4-scout-17b-16e-instruct, and more).

Sign in