AI Models

ai.KMITL provides access to multiple state-of-the-art AI models from different providers. Each model has unique strengths and characteristics. This guide will help you choose the right model for your task.

Keep Context Focused

Context limits depend on the selected model. Large files and long conversations use more quota on every follow-up; attach only what the task needs.

Premium Usage

Check the $ / $$ / $$$ price marks in the model picker, not just the premium label. Start with $ and step up only when the task needs it.

Available Models

Google Gemini & Gemma

• Gemini 2.5 Pro — Premium reasoning, vision, PDF, and effort-control support. Best for heavy research and complex drafting.
• Gemini 2.5 Flash / Flash Lite — Fast, balanced models for day-to-day chats, product copy, and product support flows.
• Gemini 2.0 Flash / Flash Lite — Stable default for most text + vision use cases with a lower usage tier.
• Gemini 2.0 Flash Lite — Lightweight fallback when you need maximum throughput.
• Google Gemma 3 (27B) — Tuned LLM for quick brainstorming with an open-source flavor.

OpenAI GPT & o-series

• GPT 5.1 / GPT 5 / GPT 5 Codex variants — Flagship premium models focused on reasoning, code generation, and multimodal use.
• GPT 4.1 family (Standard, Mini, Nano) — Balanced for structured outputs and product UX flows.
• GPT 4o & 4o Mini — Good trade-off between creativity and cost for interactive chat UIs.
• GPT OSS 20B / 120B — Open-weight variants hosted through OpenAI endpoints for experimentation.
• o3 / o3 mini / o3 pro — Reasoning-specialized models with adjustable “effort” controls; choose pro for the most rigorous analysis.
• o4 mini / o1 — Fast reasoning-first experiences when latency matters.

Anthropic Claude

• Claude Sonnet 4.5 / 4 / 3.7 / 3.5 — High-quality default for instructions, analysis, and code; Sonnet 4.5 adds optional reasoning toggle.
• Claude Haiku 4.5 — Cost-effective, fast responses, handy for support or summarization.
• Claude Opus 4 — Most capable Claude option for deep reasoning and nuanced planning.

xAI Grok

• Grok 4 — Premium reasoning + multimodal support for opinionated, creative drafting.
• Grok 4 Fast — Budget-friendly, reasoning-capable alternative suited for rapid back-and-forth chats.

DeepSeek

• DeepSeek 3.1 — Efficient reasoning model with function-calling support.
• DeepSeek R1 — Premium reasoning-first variant for algorithmic planning and math.

Meta Llama

• Llama 4 Maverick & Scout — Vision-savvy assistants ideal for creative brainstorming or design reviews.
• Llama 3.1 8B Instant — Lightweight instruct model for commodity text generation.

Image & Multimodal Creation

• GPT Image 1 — Photorealistic and illustrative assets with multiple canvas sizes.
• Gemini 2.5 Flash Image — Text-in/text-out with optional embedded generation.
• Google Nano Banana — Edit images with text.
• Google Imagen 3 & Imagen 4 — High-quality illustration.

OpenRouter

Access to various models through OpenRouter:

• DeepSeek
• Mixtral
• And more specialized models

How to Choose a Model

By Task Type

Usage tier	Start here for
$ — Cheap	Summarizing, translating, search, and everyday Q&A
$$ — Medium	Coding and analysis that need more capability
$$$ — High	Genuinely hard problems a cheaper model failed to solve

Choose a model with the capabilities your task needs, such as PDF reading or vision. Enable web search only when you need current information.

By Speed

Fastest: Groq, Gemini Flash
Balanced: Claude Sonnet, GPT-5, GPT-5.1 Slower but thorough: Claude Opus, GPT-5, o3

By Context Length

Check the selected model's limits rather than relying on a vendor's advertised maximum. A larger context window does not make a long document quota-efficient.

Switching Models

You can switch models at any time:

1. Click the model picker next to the message box, or press Ctrl + / (⌘ + / on Mac)
2. Browse, search, or filter for a model
3. Click it (or use the arrow keys and Enter) to select it
4. Continue your conversation

Finding a Model

• Sections: Favorites, Recommended (picked by ai.KMITL), and Recent sit at the top, followed by one section per provider. The icon rail on the left (a chip row on mobile) jumps straight to a section.
• Search matches model names, providers, descriptions, and uses. Typing code finds coding models first.
• Filters: narrow the list by capability (Reasoning, Vision, PDF, Tools) and by usage tier ($, $$, $$$). Capability filters combine, so Vision + PDF shows models that can read both.
• Older versions: superseded models are tucked under Older versions in each provider section. Searching or filtering shows them alongside the rest.
• Each row shows a one-line description, what the model is best for, its usage tier, and icons for what it can do.

Favorites and Your Default Model

• Click the star on any model to pin it to Favorites. Favorites are saved to your account and follow you across devices.
• Open a model's info card (the ⓘ button) and choose Set as default. Your default is the model a new device or browser starts with; after that, the picker remembers your last choice on each device.

Model Details

The info card lists what the model is best for, what it can read (images, PDFs) and do (reasoning, tools), and roughly how much of your weekly quota a typical message uses with it.

Attachments

If you've attached an image or PDF, models that can't read it are greyed out with the reason, so you can't pick one that would ignore your file.

If your weekly quota is running low while a $$$ model is selected, the picker suggests a lighter model from Recommended or your Favorites that can still read your attachments, and you can switch in one click.

Model Memory

When you switch models, the new model receives the entire conversation history, including attached files. Switching alone does not remove the cost of a long thread. For large documents, use the digest workflow in a new chat.

Model Capabilities

Text Generation

All models can generate text, but with different styles:

• Claude: Natural, conversational, detailed
• GPT: Structured, analytical, clear
• Gemini: Factual, comprehensive, thorough

Code Understanding

Best models for coding:

1. Claude Sonnet - Excellent explanations
2. GPT - Strong debugging
3. Claude Opus - Complex algorithms

Reasoning

Best models for complex reasoning:

1. Claude Opus - Deep analysis
2. GPT - Structured thinking
3. Gemini Pro - Comprehensive evaluation

Multimodal (Images, Files)

All major models support:

• ✅ Image analysis
• ✅ PDF reading
• ✅ Document understanding
• ✅ Code in images

Special Features

Claude Extended Thinking

Some Claude models support extended reasoning for complex problems. The model will "think" through the problem step by step.

GPT-4 Vision

GPT-4 models have strong vision capabilities for analyzing images, diagrams, and screenshots.

Working with Long Documents

Attach only relevant pages. For premium synthesis of a large document, extract targeted notes with a $ model first, then attach only those notes in a new chat. Summaries lose detail, so use original pages when exact wording matters.

Usage Tips

When to Use Each Model

Starting a new topic?

Start with a $ model, such as the default Flash Lite, and a focused prompt.

Need it fast?

Use Claude Haiku or Gemini Flash (speed optimized)

Complex problem?

Try a $$ model for coding or analysis; use $$$ only if a cheaper model cannot solve it.

Long document?

Use a model that supports your file type, attach only relevant pages, and consider the digest workflow before using a premium model.

Simple question?

Use Haiku or Gemini Flash Lite (quick and efficient)

Model Comparison Examples

Question: "What is photosynthesis?"

• Claude Sonnet: Detailed, educational explanation
• GPT-5: Structured, clear breakdown
• Gemini Flash: Quick, accurate summary
• Haiku: Concise, efficient answer

All correct, different styles!

Quota Considerations

Usage-Based Quota

Built-in models deduct the provider's reported cost, using the daily allowance before the weekly quota. Work billed before a response fails still counts; requests that fail before any model step runs or report no cost are not charged. See Usage Quotas.

Understanding the Tiers

The picker shows $ (Cheap), $$ (Medium), and $$$ (High). These indicate relative usage and set the preflight reserve, not a fixed charge per message. Actual usage also depends on input length, output length, reasoning, and tools. Even two models in the same tier can use different amounts of quota.

Quota-Friendly Strategies

1. Start with $ for simple tasks; speed alone is not a usage tier.
2. Step up only when needed: select a more capable model, then use Retry on an unsatisfactory answer.
3. Keep context small: switching models in a long chat keeps its history and files.
4. Use low reasoning effort for simple tasks when the model offers it.

See Making Your Quota Last for prompt, file, and continuation tips.

Bring Your Own Key (BYOK)

If you have your own API keys:

• Use models supported by your own key, subject to your provider's limits
• No platform quota deduction
• Full control over costs
• See BYOK Guide for setup

Custom Models

If you've added your own API keys, you can also

• Use beta models
• Access newest releases immediately
• Configure custom parameters

Frequently Asked Questions

Can I use multiple models in one conversation?

Yes! Switch models anytime. The conversation history transfers over.

Which model is best?

Start with a $ model, such as the default Flash Lite. Use $$ for coding or analysis that needs more capability, and $$$ for hard problems a cheaper model failed to solve.

Do different models cost different amounts?

• Yes. Usage is based on provider-reported cost, not a fixed message count.
• The picker's $, $$, and $$$ marks show relative usage; long inputs, long answers, and reasoning can increase it.
• With BYOK, your provider's pricing and limits apply instead of the included platform quota.

Why does model X sometimes give better answers than model Y?

Each model has different training, strengths, and characteristics. Try a few to find what works best for your needs.

Can I request new models?

Yes! Contact support with model requests.

---

Experiment!

Don't be afraid to try different models. You'll quickly learn which ones work best for your specific needs.

Next Steps

• Learn about Chat Interface features
• Enable Web Search & Tools
• Read Writing Great Prompts