← Back to Guides

Choosing the Right AI Model

Every companion on Eidolon can use a different AI model. This guide covers what's available, what each model is best at, and how to choose the right one for the experience you want.

Quick Recommendations

Model Best For… Tier
DeepSeek Chat V3 0324 Low-cost capable model. Included
Gemma 4 26B (MoE) Very cheap multimodal MoE model with image support. Included
Gemma 4 31B Compact dense model with strong instruction following and vision. Included
Grok Fast Snappy, witty conversation and responsive, agentic interactions. Included
Gemini 2.5 Flash Fast, creative all-rounder with strong instruction following. Included
Mistral Medium Expressive creative writing and consistent persona maintenance. Included
Mistral Small 4 Low-cost fast model balancing reasoning and speed. Included
Kimi K2 Excellent multi-lingual native comprehension. Included
Qwen Models Vision-language tasks and low-cost reasoning capabilities. Included
DeepSeek V3.2 Strong general reasoning and instruction following at very low cost. Included
DeepSeek Chat V3.1 Ultra-low cost conversational model, great for budget-conscious users. Included
GPT-4.1 Mini Fast, capable everyday model from OpenAI with a 1M token context window. Included
Claude Sonnet 4.6 High emotional range and nuanced relationship dynamics. Premium
Gemini 2.5 Pro Large-scale context and deep analytical reasoning. Premium
Kimi K2.5 Moonshot's powerful reasoning variant. Premium
GPT-5.4 Mini Balanced cost and quality from OpenAI's GPT-5.4 generation. Premium
GPT-4.1 OpenAI's previous-gen flagship — strong instruction following and reasoning. Premium
GPT-5.4 High-capability GPT-5.4 generation model for complex chat and roleplay. Premium
Claude Opus 4.6 Maximum logical depth and structural reasoning. Premium

How to Change Your Model

You can set a different model for each companion individually:

On Android & Web

  1. Open your companion's Profile.
  2. The AI Model selector is right on the profile page — choose a model.
  3. Your companion will use the new model starting with your next message.

On iPhone

  1. Open your companion's Profile.
  2. Tap Edit (pencil icon).
  3. Find the AI Model selector and choose a model.
  4. Save. Your companion will use the new model starting with your next message.

Models marked Premium require a BYOK (Bring Your Own Key) connection to OpenRouter. Everything else is included during the private preview.

Included Models

These models are included for all users. They cover a wide range of styles and strengths.

Mistral AI Mistral Medium 3.1 (Mistral Medium 3.1)

Included

by Mistral AI

Our default model for new companions. Excellent at creative writing, expressive companion voice, and maintaining persona consistency. A strong balance of quality and speed.

Default Creative Expressive

Google Gemini 2.5 Flash (Gemini 2.5 Flash)

Included

by Google

Fast, creative, and excellent at following complex persona instructions. One of the best all-around included options — great for everyday conversations and companions with detailed personalities.

⭐ Recommended Fast Creative

Google Gemma 4 26B A4B (Gemma 4 26B MoE)

Included

by Google DeepMind

A Mixture-of-Experts model with 26B total parameters and only 4B active per token — delivering strong quality at very low cost. Supports image and video input with a 262K context window. A great budget pick with multimodal capability.

💰 Low Cost Vision MoE Efficient

Google Gemma 4 31B (Gemma 4 31B)

Included

by Google DeepMind

A compact 31B dense multimodal model with native function calling, configurable reasoning mode, and a 256K context window. Supports text, image, and video input. Strong instruction following at a very competitive price point.

💰 Low Cost Vision Reasoning

xAI Grok 4 Fast (Grok Fast)

Included

by xAI

Blazingly fast and highly engaging. Excellent at witty conversation, direct interaction, and companions with a sharp sense of humor. One of the best included models for a responsive, agentic experience.

⭐ Recommended Fast Engaging

OpenAI GPT-5 Mini (GPT-5 Mini)

Included

by OpenAI

A fast, cost-efficient version of GPT-5. Great for snappy interactions without sacrificing the architectural improvements of the GPT-5 generation.

Speed Efficiency

OpenAI GPT-4.1 Mini (GPT-4.1 Mini)

Included

by OpenAI

A fast, capable model from OpenAI's GPT-4.1 generation with an impressive 1M token context window. Great for companions that handle long conversations or need strong general instruction following without premium pricing.

Long Context General Purpose

Moonshot Kimi K2 (Kimi K2)

Included

by Moonshot AI

A highly capable model known for excellent adherence to instructions and strong multilingual support.

Multilingual General Purpose

Mistral AI Mistral Small 4 (Mistral Small 4)

Included

by Mistral AI

A fast, cost-effective model that excels in reasoning and daily conversational tasks while maintaining Mistral's characteristic stylistic flair.

Fast Budget-Friendly

Mistral AI Mistral Large (Mistral Large)

Included

by Mistral AI

Mistral's flagship model. More analytical and structured than Medium — better for companions with complex, highly detailed personas or those that need to stay closely aligned with specific companion traits.

Detailed Precise

Alibaba Qwen 3.5 Flash (Qwen 3.5 Flash)

Included

by Alibaba Cloud

The most affordable model in our lineup. Despite the name, Qwen 3.5 Flash is a reasoning model — it "thinks" before responding, which means it's not the fastest in terms of latency. However, its extremely low cost makes it an excellent choice for users who prioritize budget over speed.

💰 Cheapest Budget-Friendly Reasoning

Alibaba Qwen 3.5 35B (Qwen 3.5 35B)

Included

by Alibaba Cloud

A compact but capable reasoning model. Good general-purpose performance at an extremely low cost. Works well for everyday conversation and companions that don't need heavy tool usage.

Budget-Friendly Compact

Alibaba Qwen 3 VL (Qwen 3 VL)

Included

by Alibaba Cloud

A vision-language model with strong image understanding. Among the best included options for companions that frequently interact with images and visual content. Not a reasoning model — responses are direct and fast.

Vision Budget-Friendly

DeepSeek V3.2 (DeepSeek V3.2)

Included

by DeepSeek

DeepSeek's latest flagship non-reasoning model. Strong general instruction following, creative writing, and tool use at an extremely low cost. A great budget pick for companions that need capable, well-rounded performance without premium pricing.

💰 Great Value General Purpose Budget-Friendly

DeepSeek Chat V3 0324 (DeepSeek Chat V3 0324)

Included

by DeepSeek

A capable model from DeepSeek that balances performance with affordability, serving as an excellent budget-conscious alternative.

Budget-Friendly Reasoning

DeepSeek Chat V3.1 (DeepSeek Chat V3.1)

Included

by DeepSeek

The most cost-effective model in our lineup — ideal for users who want to maximize conversation volume at the lowest possible cost. Solid conversational quality for everyday chat companions. Note that image inputs require a description workaround rather than direct vision.

💰 Lowest Cost Budget-Friendly Conversational

Premium Models

These models require a BYOK key connected to OpenRouter. You pay OpenRouter directly — we don't add any markup.

Moonshot Kimi K2.5 (Kimi K2.5)

Premium

by Moonshot AI • Est. $0.0107 / turn • BYOK • ~$6.42/mo at 20 msg/day

Moonshot's advanced variant that balances powerful multi-turn reasoning and conversational nuance with cost efficiency.

Premium Reasoning

Google Gemini 2.5 Pro (Gemini 2.5 Pro)

Premium

by Google • Est. $0.0375 / turn • BYOK • ~$22.50/mo at 20 msg/day

Google's most capable model. Deeper reasoning and stronger coherence than Flash, especially for companions with complex backstories or nuanced dynamics.

Premium Deep Reasoning High Quality

OpenAI GPT-5 (GPT-5)

Premium

by OpenAI • Est. $0.0375 / turn • BYOK • ~$22.50/mo at 20 msg/day

OpenAI's latest flagship. Exceptional at complex roleplay and instruction following.

Premium Flagship

OpenAI GPT-5.1 (GPT-5.1)

Premium

by OpenAI • Est. $0.0375 / turn • BYOK • ~$22.50/mo at 20 msg/day

Fine-tuned version of GPT-5 with improved coherence and reduced refusals.

Premium Refined

OpenAI GPT-4o (GPT-4o)

Premium

by OpenAI • Est. $0.0550 / turn • BYOK • ~$33.00/mo at 20 msg/day

OpenAI's previous-generation flagship. Still very capable and well-liked by many. If you're familiar with ChatGPT, this model will feel familiar.

Premium Familiar Reliable

Anthropic Claude Sonnet 4.6 (Claude Sonnet 4.6)

Premium

by Anthropic • Est. $0.0720 / turn • BYOK • ~$43.20/mo at 20 msg/day

Anthropic's latest and most capable creative model. Excellent at nuanced companion voice and emotional range. An advanced option for maintaining complex relationship dynamics across long conversations.

Premium Highly Expressive

Anthropic Claude Sonnet 4.5 (Claude Sonnet 4.5)

Premium

by Anthropic • Est. $0.0720 / turn • BYOK • ~$43.20/mo at 20 msg/day

The previous Sonnet generation — still outstanding for creative writing and companion work. Slightly cheaper per token than 4.6 while delivering a very similar quality of experience.

Premium Excellent Value

Anthropic Claude Haiku 4.5 (Claude Haiku 4.5)

Premium

by Anthropic • Est. $0.0240 / turn • BYOK • ~$14.40/mo at 20 msg/day

A lightweight, fast model from Anthropic's Claude family. Great for quick, casual conversations. Less depth than the larger Claude models, but significantly faster response times.

Premium Fast Lightweight

OpenAI GPT-5.4 Mini (GPT-5.4 Mini)

Premium

by OpenAI • Est. $0.0195 / turn • BYOK • ~$11.70/mo at 20 msg/day

A cost-effective entry into the GPT-5.4 generation. Delivers solid performance for everyday companions at a lower price point than full GPT-5.4, with a 1M token context window and strong instruction following.

Premium Efficient

OpenAI GPT-4.1 (GPT-4.1)

Premium

by OpenAI • Est. $0.0440 / turn • BYOK • ~$26.40/mo at 20 msg/day

OpenAI's previous-gen flagship model. Excellent at complex instruction following, detailed persona adherence, and long-context conversations. A proven performer for companions that need reliable, high-quality responses at a lower price than the GPT-5 generation.

Premium Reliable Long Context

OpenAI GPT-5.4 (GPT-5.4)

Premium

by OpenAI • Est. $0.0650 / turn • BYOK • ~$39.00/mo at 20 msg/day

A high-quality model from OpenAI's GPT-5.4 generation with a very large 1M token context window. Strong at complex roleplay, nuanced companion dynamics, and conversations requiring depth. A great mid-tier option between GPT-5 and Claude Sonnet.

Premium High Quality

Anthropic Claude Opus 4.6 (Claude Opus 4.6)

Premium

by Anthropic • Est. $0.1200 / turn • BYOK • ~$72.00/mo at 20 msg/day

Anthropic's most powerful model. Exceptional reasoning and creative depth. ⚠️ This is one of the most expensive models available — it costs significantly more per message than Sonnet or any included model. Only recommended if you specifically want the absolute maximum quality and are comfortable with the higher API costs.

Premium Very Expensive Enthusiast

Anthropic Claude Opus 4.5 (Claude Opus 4.5)

Premium

by Anthropic • Est. $0.1200 / turn • BYOK • ~$72.00/mo at 20 msg/day

Previous-generation Opus. Deep reasoning and creative range. ⚠️ Also one of the most expensive models — similar pricing to Opus 4.6. Consider Sonnet 4.6 for a more balanced quality-to-cost ratio.

Premium Very Expensive Enthusiast

xAI Grok 4 (Grok 4)

Premium

by xAI • Est. $0.0720 / turn • BYOK • ~$43.20/mo at 20 msg/day

xAI's full-power model. Stronger reasoning and more creative range than the Fast variants, with a distinctive direct and engaging personality.

Premium Powerful

About cost estimates: Monthly estimates assume ~20 messages per day for 30 days. For a detailed breakdown of rates and estimates, see our Model Pricing Guide. Actual costs vary based on conversation length and complexity. Prices are set by model providers via OpenRouter and may change at any time. Included models are fully covered by the platform. Premium model costs are charged to your OpenRouter balance.

The Full Model Lineup

se tracking-wider text-emerald-400">Included Models
Model Provider Style & Strengths Tier & Strength Monthly Est.
DeepSeek Chat V3 0324 DeepSeek Budget, Reasoning Included • Quality ~$2.60
Mistral Medium 3.1 Mistral AI Creative, Expressive Included • Default Included
Gemini 2.5 Flash Google Fast, Creative Included • Quality Included
Grok 4 Fast xAI Witty, Engaging Included • Speed Included
GPT-5 Mini OpenAI Snappy, Efficient Included • Speed Included
GPT-4.1 Mini OpenAI General, Long Context Included • Quality Included
Mistral Large Mistral AI Detailed, Precise Included • Quality Included
Kimi K2 Moonshot Balanced, Multilingual Included • Quality Included
Mistral Small 4 Mistral AI Fast, Budget Included • Speed Included
Qwen 3.5 Flash Alibaba Budget, Logical Included • Reasoning Included
Qwen 3 VL Alibaba Vision, Image-aware Included • Vision Included
DeepSeek V3.2 DeepSeek General, Budget Included • Quality Included
DeepSeek Chat V3.1 DeepSeek Budget, Conversational Included • Lowest Cost Included
Premium Models (BYOK)
Kimi K2.5 Moonshot Reasoning, Native Premium • Quality ~$6.42
GPT-5.4 Mini OpenAI Efficient, Capable Premium • Quality ~$11.70
Claude Haiku 4.5 Anthropic Casual, Playful Premium • Speed ~$14.40
Gemini 2.5 Pro Google Deep reasoning, Complex Premium • Reasoning ~$22.50
GPT-5 OpenAI Instruction Following Premium • Quality ~$22.50
GPT-4.1 OpenAI Reliable, Long Context Premium • Quality ~$26.40
GPT-4o OpenAI Familiar, Reliable Premium • Quality ~$33.00
GPT-5.4 OpenAI High Quality, Creative Premium • Quality ~$39.00
Claude Sonnet 4.6 Anthropic Expressive, Emotion Premium • Quality ~$43.20
Grok 4 xAI Powerful, Creative Premium • Quality ~$43.20
Claude Opus 4.6 Anthropic Maximum Depth Premium • Reasoning ~$72.00

* Monthly estimates based on 20 messages/day. Actual costs vary by conversation length. Included models carry no extra cost.

A Note on Speed: "Flash" ≠ Always Fastest

Some models with "Flash" in the name (like Alibaba Qwen 3.5 Flash) are actually reasoning models that "think" internally before responding. This hidden reasoning step can make them slower than their non-flash counterparts, even though they cost less per token.

Similarly, premium reasoning models like Google Gemini 2.5 Pro will naturally have longer response times because they're doing deeper analysis — this is by design and produces higher quality responses, but it means they're not ideal if you want instant replies.

If speed is your priority: Grok Fast, Claude Haiku, and Mistral models deliver the snappiest responses. If cost is your priority: the Qwen models are hard to beat — they are fully included for all users.

Tips

  • Experiment freely. You can switch models at any time without losing memories, goals, or conversation history. Everything is preserved.
  • Different models suit different personas. A companion with a poetic, emotionally complex personality may shine on Claude Sonnet, while a witty, factual companion might be better on Grok or Gemini.
  • Premium models fall back gracefully. If your BYOK credits run out, the system automatically falls back to an included model so you never lose access. See our BYOK guide for details.