How to Use Multiple Current AI Models on Magicdoor

Practical current-model workflows for drafting, research and files, using registry-backed rates and avoiding retired picker models or guaranteed savings.

Last updated: October 7, 2026.

Magicdoor lets you switch among currently supported chat models within a conversation. Pick each step based on its file and tool needs, then verify the output; there is no universal quality ranking or measured savings percentage.

Compare a few current options

ModelStandard input / output per 1M tokensA task to try
GPT-6 Luna$0.10 / $0.50A low-rate draft or compatible-file code interpreter task
Claude Haiku 5.5$0.10 / $0.50Search-enabled question, PDF or image input
Claude Sonnet 5.5$2.00 / $10.00Compare revisions of your own writing
GPT-6.2 Sol$2.00 / $10.00Compare a harder analysis or code task
Claude Opus 5.5$4.00 / $20.00Subscriber-only second pass

Rates exclude cache, long-context and relevant tool adjustments. Read the model cost guide.

Three useful workflows

Draft and revise

Ask Luna for an initial outline, then compare Sonnet or Sol on one section. Edit the final draft yourself and check any references. Do not assume the higher-rate model writes better for every brief.

Find sources and analyze

Select a search-enabled model such as Claude Haiku 5.5 or Sol. Ask for original URLs, open them and verify each claim. Then switch to another current model if you want a separate synthesis of those verified sources. Old Perplexity Reasoning and Deep Research (pplx) picker models are retired, and Auto does not promise a Perplexity search pass.

Work with files

Check your model's compatibility for the actual file type. Luna and Sol support code interpreter for compatible files, while Claude Haiku 5.5 supports PDF and vision but not Magicdoor's code interpreter. Compare the answer with the file contents and rerun calculations outside the model where needed.

Basic costs $6/month with $2 included usage balance in the first month and $1 on renewal; Plus costs $12/month with $7 monthly usage balance. Additional usage is billed separately. Compare your own workload before assuming a particular total cost.

Try a multi-model workflow

FAQ

Why switch models during a task?

Different current models support different tools and rates. Compare their outputs on a representative task rather than assuming one model is always better.

Which lower-rate Magicdoor model can I start with?

GPT-6 Luna has standard rates of $0.10 input and $0.50 output per million tokens. Claude Haiku 5.5 costs $0.10 and $0.50. Actual charges vary with context, caching and tools.

Does Magicdoor guarantee savings from model switching?

No. Compare the subscription price, included usage balance and your actual usage with alternatives. A lower token rate is not a guarantee of answer quality or total savings.