Cost per model
Understanding Model Costs
Different AI models have varying pricing structures based on their capabilities and computational requirements. Here's a detailed breakdown of the costs for each available model.
The good news: AI chat is surprisingly cheap. It's quite difficult to spend more than a few cents per conversation, even with the most capable models.
What is Blended Cost?
Blended cost is a simplified metric that represents the average cost per 1 million tokens for a model, taking into account typical usage patterns where conversations include both input (prompt) and output (completion) tokens. This gives you a single number to compare models at a glance.
The blended cost is calculated assuming output tokens are roughly 3x input tokens — for every word you send, you get about three words back. This reflects real-world usage patterns.
Real cost examples
A short question for Perplexity Reasoning:
- ~400 total tokens
- low token cost plus a small web-search request fee
- Most users stay within the $1 included credit for months
A longer chat with Claude Sonnet 5 on a complex topic:
- ~3,000 total tokens
- ~$0.05 total cost
Using GPT-5.6 Luna for quick tasks all day:
- Even 100 back-and-forth messages costs less than $0.10
Complete Model Pricing Table
The table below shows all available chat models with their pricing. The Blended Cost column is a quick comparison metric for typical usage.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Blended Cost | Additional Fees |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | $0.14 | $0.28 | $0.21 | Cached input: $0.03 |
| GLM-5.3 Flash | $0.15 | $0.50 | $0.24 | Cached input: $0.03 |
| GPT-5.6 Luna | $0.20 | $1.20 | $0.40 | Cached input: $0.02 |
| MiniMax M3 | $0.30 | $1.20 | $0.53 | Cached input: $0.06 |
| GLM-5.3 | $1.40 | $4.40 | $2.15 | Cached input: $0.26 |
| Gemini 3.7 Flash | $0.75 | $3.75 | $2.25 | Cached input: $0.075 |
| Grok 4.6 | $2.00* | $6.00* | $2.80* | Cached input: $0.50* |
| Qwen3.8 Max | $2.50 | $6.25 | $3.30 | Cached input: $0.50 |
| Claude Sonnet 5 | $2.00 | $10.00 | $4.00 | Cached input: $0.20 |
| GPT-5.6 Sol | $4.00* | $20.00* | $8.00* | Cached input: $0.40* |
| Kimi K3 | $3.00 | $15.00 | $9.00 | Cached input: $0.30 |
| Claude Opus 5 | $5.00 | $25.00 | $10.00 | Cached input: $0.50 |
| Claude Fable 5.1 | $10.00 | $50.00 | $20.00 | Cached input: $0.25 |
Understanding the Table
- Input/Output costs are per 1 million tokens
- Blended Cost represents typical usage combining input and output at a 1:3 ratio
- Models are sorted roughly by cost tier
* Grok 4.6 prices shown apply below 200K prompt tokens. At 200K prompt tokens or more, input costs $4.00, cached input costs $1.00, output costs $12.00, and the 80% input / 20% output blended comparison becomes $5.60. GPT-5.6 Sol also has higher long-context pricing above its 272K threshold.
Image Model Pricing
Image generation is priced per image, not per token.
| Model | Cost per image | Notes |
|---|---|---|
| Recraft Upscaler | $0.006 | Upscaling only |
| Seedream 5 Pro | $0.045 | 1K output, up to 10 reference images |
| Google Nano Banana 2 | $0.067 | Fast and affordable |
| Recraft V4.1 | $0.04 | Illustration and design-focused |
| Flux 2 Pro | $0.03 | Excellent for realistic photography |
| Google Nano Banana Pro (2K) | $0.15 | 2K resolution output |
| GPT Image 2.5 Flare | $0.012-$0.50 (Medium default: $0.047) | Faster everyday generation and editing with up to 4 images |
| GPT Image 2.5 Sunburst | $0.012-$0.50 (High default: $0.128) | Precision-focused generation and editing with up to 4 images |
Active image generation costs $0.012–$0.50 per image. At $6/month including a $1 usage credit, that's roughly 2–83 images on the house before you need to top up.
Cost Optimization Tips
- Right-size your model: Use GLM-5.3 Flash or GPT-5.6 Luna for quick tasks. Save Claude Opus 5 or GPT-5.6 Sol for when the task actually needs a premium model.
- Keep conversations short: With every message you send, the entire conversation history is sent to the model. Long threads get expensive fast. Start fresh chats for new topics.
- Use the model switcher: Spotted mid-conversation that a cheaper model would do? Switch — magicdoor.ai lets you change models without losing context.
- Watch the live cost indicator: The UI shows real-time cost per message so you always know where you stand.
What does $6/month actually get you?
Your subscription includes a $1 usage credit. For most users, that covers months of normal use. When it runs out, top up your balance (it never expires) and pay only for what you use. Around 70% of subscribers never need to top up at all.
For more help choosing the right model for your task, see the model selection guide.
FAQ
How is AI pricing calculated?
Most AI models charge per token—specifically per 1 million tokens for input and output. A token is roughly 4 characters or 3/4 of a word. A typical email might be 200-500 tokens.
What does "blended cost" mean?
Blended cost averages input and output pricing assuming you write 1 word and get back 3 words. It's a quick comparison metric—lower blended cost means cheaper typical usage.
How can I reduce my AI costs?
Use budget models like GLM-5.3 Flash or GPT-5.6 Luna for simple tasks. Start new chats for new topics (long conversations cost more). Save premium models for tasks that actually need them.
Does the $1 credit expire?
No. Your credit never expires. You can top up whenever you want, and added balance also never expires.
How much does image generation cost?
Image generation is priced per image, not per token. Active generation models cost $0.012–$0.50 per image. Your $1 credit covers roughly 2–83 images.
Related Resources
Best AI for Coding in 2026: A Practical Model Guide
A practical comparison of AI models for coding in 2026, covering complex refactors, debugging, code review, boilerplate, and explaining code — with current pricing and starting points to test.
AI Cost Optimization: A Practical Model Routing and Budget Guide
A decision framework for controlling AI costs with current chat and image models, measured usage, model escalation rules, and an honest flat-rate break-even check.
DeepSeek V4.1 Flash Guide - Context, Privacy, and Pricing
A practical guide to DeepSeek V4.1 Flash on magicdoor.ai, including its 1,048,576-token context, tools, pricing, and enforced zero-data-retention routing.
GPT Image 2.5 Flare and Sunburst Guide on magicdoor.ai
Complete guide to GPT Image 2.5 Flare and Sunburst on magicdoor.ai, including current pricing, editing support, aspect ratios, and when to choose each OpenAI image model.