Providers Overview

Tarsk supports 29 model providers. Each provider requires its own API key (or OAuth for some). You can enable multiple providers simultaneously and switch between models per thread.

Direct Providers

These providers host their own models. Bring an API key from the provider’s website.

Anthropic

Models: Claude Opus 4.6, Claude Sonnet 4.6, Claude Haiku 4.5 Get a key: console.anthropic.com

Claude models are strong at coding, reasoning, and following complex instructions. Claude Sonnet 4 offers an excellent balance of capability and speed.

OpenAI

Models: GPT-5, GPT-4.1, GPT-4o, GPT-4, o3-mini, o4-mini, Codex series
Get a key: platform.openai.com

Google

Models: Gemini 3 Pro Preview, Gemini 3 Flash Preview, Gemini 2.5 Pro, Gemini 2.5 Flash
Get a key: aistudio.google.com

Gemini 2.5 Flash is a fast, cost-effective option with a large context window.

xAI

Models: Grok-4, Grok-4-1-Fast, Grok-Code-Fast-1
Get a key: console.x.ai

DeepSeek

Models: DeepSeek Chat (DeepSeek-V3), DeepSeek Reasoner (R1)
Get a key: platform.deepseek.com

DeepSeek R1 is a strong reasoning model. DeepSeek V3 is competitive with frontier models for coding tasks.

Groq

Models: Kimi K2 Instruct, GPT-OSS 120B
Get a key: console.groq.com

Groq provides very fast inference via custom hardware.

Cerebras

Models: GPT-OSS 120B, ZAI-GLM-4.7, Qwen-3-235B
Get a key: cloud.cerebras.ai

MoonshotAI

Models: Kimi K2.5, Kimi K2 Thinking
Variants: moonshotai (international), moonshotai-cn (China endpoint)
Get a key: platform.moonshot.cn

MiniMax

Models: MiniMax M2, M2.1, M2.5
Variants: minimax (international), minimax-cn (China endpoint)

ZhipuAI

Models: GLM-4.5 through GLM-4.7
Variants: zhipuai, zhipuai-coding-plan

Qwen

Models: Qwen Coder
Get a key: dashscope.aliyuncs.com

NVIDIA

Models: MiniMax M2.1, Kimi K2, GPT-OSS 120B, Qwen3 Coder
Get a key: build.nvidia.com

HuggingFace

Models: Qwen3 Coder, Kimi K2.5, GLM-4.7
Get a key: huggingface.co/settings/tokens

VolcEngine

Models: DeepSeek V3.1, Doubao Seed
Baidu/ByteDance cloud inference platform

Xiaomi

Models: Mimo V2 Flash

---

Aggregator Providers

Aggregators route requests to multiple underlying models through a single API key. They are useful for accessing many models without managing separate keys.

OpenRouter

Models: Hundreds of models from Anthropic, OpenAI, Google, Meta, and more
Get a key: openrouter.ai
Credits: Balance visible in Settings

OpenRouter is the most comprehensive aggregator. A single key accesses almost every major model. Some models are free; paid models are charged per-token. See the dedicated OpenRouter guide for setup details.

AIHubMix

Models: DeepSeek, Claude, Kimi, and more via a single API
Get a key: aihubmix.com
Credits: Balance visible in Settings

SiliconFlow

Models: Qwen, DeepSeek, Kimi
Variants: siliconflow (international), siliconflow-cn (China endpoint)

ModelScope

Models: Qwen, GLM, DeepSeek

iFlow

Models: DeepSeek R1, DeepSeek V3, GLM series, Kimi, Qwen3

CanopyWave

Models: DeepSeek, MiniMax, Kimi, GPT-OSS

OpenCode

Models: Multi-provider aggregator

ModelWatch

Models: Multi-provider aggregator

ZenMux

Models: Multi-provider aggregator

Poe

Models: Claude, GPT, Gemini, Grok (via Poe subscription)
Get a key: poe.com

---

OAuth Providers

These providers use OAuth rather than static API keys.

GitHub Copilot

Models: Claude, Gemini, GPT-5, Grok
Auth: Connect via your GitHub account — no separate API key needed if you have a Copilot subscription

Codex

Models: GPT-5.1 Codex variants
Auth: OAuth via your OpenAI account

---

Specialised Providers

Provider	Models	Notes
zai-coding-plan	GLM-4.5 through GLM-5	Planning-focused variants
kimi-coding-plan	Kimi for Coding	Specialised coding mode
zhipuai-coding-plan	GLM coding variants	Coding + planning split

Next: Configure providers and enable models →