Model Recommendations

Not sure which model to use? This guide picks the best model for common tasks based on current provider catalogs (OpenRouter, AIHubMix, Cloudflare Workers AI). All models work through Arcana — just set the provider key and go.

Model data sourced from: OpenRouter API, AIHubMix API, Cloudflare AI. Last verified Jul 25, 2026.

Quick Picks

Task	Recommended	Context	Why
Everyday coding	`claude-sonnet-5` or `gpt-5.4`	1M	Best all-round quality and tool use
Budget coding	`gpt-4.1-mini` or `gemini-3.6-flash`	1M	Very cheap, still very capable
Deep reasoning	`claude-opus-5` or `claude-fable-5`	1M	Flagship reasoning and agentic work
Long documents	`gemini-3.6-flash` or `kimi-k3`	1M	Massive context windows
Fast responses	`gpt-4.1-nano` or `gemini-3.5-flash-lite`	1M	Lowest latency, high throughput
Free / local	`ollama/llama3`	varies	No API key, runs on your machine

By Provider

OpenAI

Model	Context	Strengths	Cost Tier
`gpt-5.6-sol`	1M	State-of-the-art coding, science, cybersecurity	High
`gpt-5.6-terra`	1M	Balanced intelligence and cost (mini tier)	Medium
`gpt-5.6-luna`	1M	Cost-sensitive, high-volume workloads (nano tier)	Low
`gpt-5.5`	varies	Strong coding, reasoning, multimodal	High
`gpt-5.4`	varies	Flagship model, strong all-round	High
`gpt-5.4-mini`	varies	Smaller, faster, cost-efficient	Medium
`gpt-5.4-nano`	varies	Smallest and fastest, edge use cases	Low
`gpt-5`	varies	Coding, writing, reasoning	High
`gpt-5-mini`	varies	Lightweight, high-volume tasks	Medium
`gpt-5-nano`	varies	Smallest GPT-5, low latency	Low
`gpt-4.1`	1M	Reliable flagship, excellent tool use	Medium
`gpt-4.1-mini`	1M	Fast, affordable, still very capable	Low
`gpt-4.1-nano`	1M	Smallest, cheapest, high-throughput	Very Low

Anthropic

Model	Context	Strengths	Cost Tier
`claude-fable-5`	1M	Most capable model, demanding reasoning, long-horizon agentic work	Very High
`claude-opus-5`	1M	Flagship for complex agentic coding and enterprise work	High
`claude-opus-4.8`	200K	Most powerful publicly available, complex reasoning	High
`claude-opus-4.7`	1M	Step-change improvement in agentic coding	High
`claude-sonnet-5`	1M	Most agentic Sonnet, coding, tool use, reasoning	Medium
`claude-sonnet-4.6`	varies	Balanced coding, reasoning, agentic capabilities	Medium
`claude-haiku-4.5`	varies	Coding performance at 1/3 cost, 2x speed	Low

Google Gemini

Model	Context	Strengths	Cost Tier
`gemini-3.6-flash`	1M	Frontier intelligence, code generation, agentic execution, spatial reasoning	Low
`gemini-3.5-flash`	1M	Fast multimodal, sub-agent deployment, long-horizon tasks	Low
`gemini-3.5-flash-lite`	1M	Lowest latency, high-throughput subagent tasks	Very Low
`gemini-3.1-pro`	1M	Most intelligent Gemini, improved reasoning	Medium
`gemini-2.5-pro`	1M	Strong reasoning, thinking support	Medium
`gemini-2.5-flash`	1M	Fast multimodal, strong reasoning	Low

xAI (Grok)

Model	Context	Strengths	Cost Tier
`grok-4.5`	1M	Coding, science, engineering, math — exceeds comparable frontier models	Medium
`grok-build-0.1`	256K	Fast coding model for agentic coding workflows	Low

MoonshotAI (Kimi)

Model	Context	Strengths	Cost Tier
`kimi-k3`	1M	2.8T params, long-horizon coding, knowledge work, #1 in multiple arenas	Medium
`coding-kimi-k3`	1M	Coding-specialized variant of Kimi K3	Low

DeepSeek

Model	Context	Strengths	Cost Tier
`deepseek-v4-pro`	varies	High-capability reasoning, production-grade inference	Low

Alibaba (Qwen)

Model	Context	Strengths	Cost Tier
`qwen3.8-max-preview`	~1M	2.4T params, full-stack dev, data analysis, Office workflows	Low
`qwen3-max`	varies	Strong coding, reasoning, multilingual	Low

Other Notable Models

Model	Provider	Context	Strengths
`glm-5.2`	Zhipu AI	1M	Project-level engineering, full dev workflow
`muse-spark-1.1`	Meta	1M	Multimodal reasoning, agentic tasks, accepts video/audio/PDF
`hy3`	Tencent	256K	295B MoE, coding agent, long-doc analysis
`m3`	MiniMax	1M	Frontier coding, agentic capabilities
`doubao-seed-2.1-pro`	ByteDance	256K	Enterprise coding, agent, multimodal
`laguna-s-2.1`	Poolside	1M	118B coding agent model, 70.2% Terminal-Bench

By Task Type

Coding & Development

Model	Strengths	Cost
`claude-opus-5`	End-to-end software tasks, code review, bug finding, agentic coding	High
`gpt-5.6-sol`	State-of-the-art across coding, knowledge, science	High
`claude-sonnet-5`	Strong coding, tool use, follows instructions precisely	Medium
`kimi-k3`	#1 in coding arenas, 1M context, long-horizon coding	Medium
`gpt-4.1`	Reliable code generation, debugging, refactoring	Medium
`deepseek-v4-pro`	High-capability reasoning, very cheap	Very Low
`laguna-s-2.1`	Purpose-built coding agent, 1M context	Low

Writing & Content

Model	Strengths	Cost
`claude-sonnet-5`	Natural prose, creative writing, editing, professional work	Medium
`gpt-5.4`	Structured content, technical writing, strong all-round	High
`gemini-3.6-flash`	Long-form content, document analysis, 1M context	Low

Analysis & Research

Model	Strengths	Cost
`claude-fable-5`	Most capable reasoning, 1M context, nuanced analysis	Very High
`gemini-3.6-flash`	1M context, analyze entire codebases or document sets	Low
`kimi-k3`	2.8T params, #1 in knowledge work arenas	Medium

Creative & Brainstorming

Model	Strengths	Cost
`claude-sonnet-5`	Creative writing, worldbuilding, dialogue	Medium
`gpt-5.4`	Brainstorming, diverse ideas, multimodal	High
`grok-4.5`	Real-time context, current events, humor	Medium

The Utility Model

Arcana uses a secondary "utility model" for background tasks like fact extraction, session compaction, and summarization. Set this to a cheap, fast model:

{
  "model": "claude-sonnet-5",
  "utilityModel": "gpt-4.1-nano"
}

Good utility model picks: gpt-4.1-nano, gpt-4.1-mini, gemini-3.5-flash-lite, gemini-3.6-flash.

Switching Models

You can switch models mid-session without restarting:

# In the TUI, type:
/providers

# Or set in config:
export ARCANA_MODEL=claude-sonnet-5

Tips

Start with claude-sonnet-5 or gpt-5.4 — they handle most tasks well
Use gpt-4.1-nano or gemini-3.5-flash-lite as the utility model to keep costs low
Switch to claude-opus-5 or claude-fable-5 only when you need deep reasoning
For budget coding, deepseek-v4-pro and kimi-k3 punch well above their price
Run arcana models to see all available models from your detected providers
Prices change frequently — check provider pricing pages for current rates

Last updated: Jul 25, 2026 · Data from OpenRouter, AIHubMix, Cloudflare AI