Update (October 23, 2026): xiaomi-mimo-v2.5-pro-free was retired on 21 October 2026 and now returns 404 model_retired. If a saved configuration still uses it, switch the Model field to xiaomi-mimo-v2.6-pro-free.
Short answer: start with xiaomi-mimo-v2.6-pro-free, save one or two general chat models as backups, and keep low-cap and coding models out of long sessions. Treat that as a starting point, not a ranking. There's no official roleplay benchmark for AIHubMix's free models, and none of the model pages mention roleplay. This post covers how to choose and how to test, so you can decide based on your own characters.
Why there's no "best model" list
AIHubMix's free model catalog lists dozens of models at $0. Each model page covers context window, capabilities, and limits. None of them say anything about roleplay or character writing, and how well a model plays a character depends on the bot, your prompts, and your own preferences.
So the useful questions are narrower. Which models are worth trying first? Which ones will cut you off halfway through a scene? Three things answer that better than model size:
- Caps: how many replies you get per day and per minute.
- Model type: a general chat model, or a version built for coding tools.
- Stability: whether the model is new, in preview, or already retired.
Starting points
| Model ID | Daily cap | Context | Why it's here |
|---|---|---|---|
| xiaomi-mimo-v2.6-pro-free | 100 | 1.05M | Top pick |
| xiaomi-mimo-v2.6-flash-free | 100 | 1.05M | Same-family backup |
| minimax-m2.7-free | 100 | 205K | General MiniMax |
| glm-4.7-flash-free | 100 | 200K | General GLM |
Caps apply after the one-time top-up; before it, every free model draws from the same 10 trial calls. The caps come from the free models table: 100 requests per day and 10 per minute for every model listed here.
MiMo V2.6 Pro is our first pick. Early roleplay feedback has been good, especially for keeping scenes moving. Its model page describes it as the free version of Xiaomi's "trillion-parameter, natively omni-modal flagship reasoning model", with a 1.05M-token context window and up to 131K output tokens.
It's worth saving a backup or two before you need them.
MiMo V2.6 Flash is the obvious backup if you like how MiMo plays your characters. It's from the same family as V2.6 Pro and has the same 1.05M context window.
MiniMax M2.7 and GLM 4.7 Flash are general chat models from two other families. Saving one of them means a busy MiMo doesn't have to end your session.
MiMo V2.5 Pro is going offline. xiaomi-mimo-v2.5-pro-free will be retired on 21 October 2026. If one of your saved configurations uses it, switch the Model field to xiaomi-mimo-v2.6-pro-free before then. After that date, requests to V2.5 Pro will return 404 model_retired.
What the "coding-" models are
More than a dozen free models start with "coding-": coding-glm-5.1-free, coding-minimax-m2.7-free, and so on. AIHubMix describes them as dedicated versions for coding tools. The coding GLM 5.3 page calls its Flash sibling "a dedicated version of GLM 5.3 Flash built for AI coding and Coding Agent workflows". It also describes the paid coding-glm-5.3 as a "limited-time preview" that is "intended for testing and evaluation only".
Nothing stops you from using them in JanitorAI, because the pages don't restrict chat. Our advice is to start with a general model (one without "coding-") and keep coding versions as extra backups. Both GLM and MiniMax have a general version on the free list: glm-4.7-flash-free and minimax-m2.7-free.
Models to keep out of long sessions
Some free models count each call as several requests, so their caps are lower:
| Model ID | Daily cap | Per minute |
|---|---|---|
| qwen3.6-plus-preview-free | 50 | 5 |
| coding-glm-5-turbo-free | 50 | 5 |
| k2.6-code-preview-free | 50 | 5 |
Every reroll in JanitorAI is a new request, so 50 requests don't last long in an active scene. These models are fine for a quick comparison but not for a long roleplay chat.
Why a 1M context window doesn't help in JanitorAI
Many free models advertise 1M tokens of context or more. In JanitorAI that number barely matters. With an outside model, Janitor's context slider runs from 0 to 128,000 tokens, so Janitor never sends more than 128K per turn. Every model in the starting list can take that.
What matters is the daily token pool. All free models share 1M tokens per day, and every turn sends the whole context again, so a large context setting drains the pool fast. Start between 16,000 and 32,000. The limits post in this series works through the numbers.
Reasoning models and max tokens
MiMo V2.6 Pro is a reasoning model, so it may think before it replies.
For a reasoning model, max tokens 0 is the safer starting point. In Janitor's settings, 0 "lets the model choose the length", while any other value is a hard stop. A low hard stop may eat into the thinking and leave a short or empty reply.
New, preview, and retired models
The free list changes, so check a model's page before you settle on it:
- Preview and limited models can be unstable. They may occasionally return 429 errors.
- Models get retired. Requests to a retired free model return 404 model_retired. MiMo V2.5 Pro is next, on 21 October 2026.
- New models can change. A model tagged New may have its limits or availability adjusted after launch.
As for DeepSeek: as of the free list's 12 October 2026 update, no DeepSeek model is on it.
A setup that survives busy hours
- Save 2–3 proxy configurations, one per model, such as "MiMo Pro", "MiMo Flash", and "MiniMax". Open a configuration that works, click Duplicate, and change only the Model field; the URL and key stay the same.
- Switch when a model returns 429 or stalls. A 429 means you've hit a cap or the model is busy. Retrying right away rarely helps, so wait a minute or move to the next configuration.
- Compare on the same scene. Use the same bot, the same opening, and the same few messages for each model, then read the replies side by side. Decide on what you see.
- Copy model IDs exactly. Copy the ID from the model page; one wrong character gives a model error.
FAQ
Which AIHubMix free model is best for JanitorAI roleplay?
There's no official ranking. We recommend starting with xiaomi-mimo-v2.6-pro-free. Keep a general model such as minimax-m2.7-free or glm-4.7-flash-free as a backup, and compare them on your own scenes.
Can I use the coding- models for roleplay?
Yes. Nothing on their pages restricts chat. They're versions built for coding tools, so a general model is the more natural first choice.
Why do some free models stop after a few replies?
Some count each call as several requests, so their caps are lower, such as 50 requests per day and 5 per minute for qwen3.6-plus-preview-free. Others in the free table have been retired and return a model_retired error.
Can I still use MiMo V2.5 Pro?
Only until 21 October 2026, when xiaomi-mimo-v2.5-pro-free goes offline. Move your configuration to xiaomi-mimo-v2.6-pro-free now.
Does a 1M-token context window give my bot a longer memory in JanitorAI?
No. Janitor's context slider stops at 128,000 tokens with an outside model, and a bigger context drains the shared 1M-token daily pool faster.
What should I do when my model returns 429?
Wait a minute or switch to another saved configuration. A 429 means you've hit a cap or the model is busy.
Is DeepSeek free on AIHubMix?
Not as of the free list's 12 October 2026 update. No DeepSeek model is on it.
What max tokens setting should I use for MiMo V2.6 Pro?
Start at 0. It lets the model choose the length, so a low hard stop doesn't cut into the thinking or the reply.
Keep reading: the AIHubMix for JanitorAI series
- Haven't connected a model yet? The setup guide covers the URL, key, and model ID, and the errors you'll see when one is wrong: How to Use AIHubMix Free Models in JanitorAI: Proxy Setup and Common Errors
- Choosing a model is easier once you know how far 100 requests and 1M tokens go at your context size: AIHubMix Free Model Limits for JanitorAI: 10 Trial Calls, the $1 Unlock, and the Token Cap
- The daily caps in this post apply only after the one-time top-up: How to Pay on AIHubMix with Alipay
- Your roleplay chats aren't stored on AIHubMix; here's how zero data retention works: AIHubMix Supports JanitorAI Roleplay Chat, with Zero Data Retention on by Default
Sources
- Free AI Models (AIHubMix)
- Xiaomi MiMo V2.6 Pro (free) (AIHubMix)
- Changelog (AIHubMix docs)
- MiniMax M2.7 (free) (AIHubMix)
- GLM 4.7 Flash (free) (AIHubMix)
