FR

The AI models on aidock

For each model: who runs it, where your message is processed, and whether it may be used to train AI. Everything is sourced and dated.

Try in the chatSee the Barometer
16models in the catalogue
11with no training on your messages
6providers checked
11offers left out, with the reason

All models

Filters look at each model’s main route: the one the chat tries first.

gpt-oss-120b

OpenAI, United StatesOpen model

Open model from OpenAI under the Apache 2.0 licence. It thinks before answering: good for reasoning, analysis and writing.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
  • GroqFree
    No trainingHosted in the US
  • Scalewayaidock budget
    No trainingHosted in France
Context
131,000 tokens
Reads images
No
Thinks before answering
Yes
Open weights
Yes
  • Reasoning
  • Writing
  • Code
See the page gpt-oss-120bTry gpt-oss-120b

gpt-oss-20b

OpenAI, United StatesOpen model

Small version of gpt-oss: faster, ideal for simple requests.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
  • GroqFree
    No trainingHosted in the US
Context
131,000 tokens
Reads images
No
Thinks before answering
Yes
Open weights
Yes
  • Speed
  • Writing
See the page gpt-oss-20bTry gpt-oss-20b

Gemma 4 26B

Google, United StatesOpen model

Open model from Google. It reads very long documents and understands images.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
  • Scalewayaidock budget
    No trainingHosted in France
Context
262,000 tokens
Reads images
Yes
Thinks before answering
Yes
Open weights
Yes
  • Summaries
  • Long documents
  • Speed
See the page Gemma 4 26BTry Gemma 4 26B

Gemma 4 31B

Google, United StatesOpen model

Larger version of Gemma 4, offered for free through OpenRouter.

Where your message goes

  • OpenRouterFree
    No trainingHosted in the US
Context
262,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Writing
  • Summaries
See the page Gemma 4 31BTry Gemma 4 31B

Qwen 3.8 27B

Alibaba (Qwen), ChinaOpen model

Open model from Alibaba’s Qwen team. Versatile, comfortable with code and long documents.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
Context
262,000 tokens
Reads images
Yes
Thinks before answering
Yes
Open weights
Yes
  • Code
  • Long documents
  • Reasoning
See the page Qwen 3.8 27BTry Qwen 3.8 27B

Llama 4 Scout

Meta, United StatesOpen model

Open model from Meta. Fast and able to read images.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
Context
131,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Speed
  • Writing
See the page Llama 4 ScoutTry Llama 4 Scout

Mistral Small 3.1

Mistral AI, FranceOpen model

Open model from Mistral AI, served here by Cloudflare. Fast and comfortable in French.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
Context
131,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Writing
  • Translation
  • Speed
See the page Mistral Small 3.1Try Mistral Small 3.1

Nemotron 3 120B

NVIDIA, United StatesOpen model

Open model from NVIDIA, built to reason and follow multi-step instructions.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
Context
262,000 tokens
Reads images
No
Thinks before answering
Yes
Open weights
Yes
  • Reasoning
  • Code
See the page Nemotron 3 120BTry Nemotron 3 120B

Mistral Small 4

Mistral AI, FranceOpen model

Recent small model from Mistral AI, a French company. Fast, very comfortable in French and translation.

Where your message goes

  • Mistral AIFree
    May trainLocation not guaranteed
Context
262,000 tokens
Reads images
Not stated
Thinks before answering
Yes
Open weights
Yes
  • Writing
  • Translation
  • Speed
See the page Mistral Small 4Try Mistral Small 4

Mistral Medium 3.5

Mistral AI, FranceOpen model

Mid-size model from Mistral AI, stronger than Small, for demanding texts.

Where your message goes

  • Mistral AIFree
    May trainLocation not guaranteed
  • Scalewayaidock budget
    No trainingHosted in France
Context
262,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Writing
  • Translation
  • Summaries
See the page Mistral Medium 3.5Try Mistral Medium 3.5

Ministral 14B

Mistral AI, FranceOpen model

Compact model from Mistral AI, fast for simple tasks.

Where your message goes

  • Mistral AIFree
    May trainLocation not guaranteed
Context
262,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Speed
See the page Ministral 14BTry Ministral 14B

Nemotron 3 Ultra

NVIDIA, United StatesOpen model

Very large model from NVIDIA, free through OpenRouter. Its provider may train its models on your messages.

Where your message goes

  • OpenRouterFree
    May trainHosted in the US
Context
1,000,000 tokens
Reads images
No
Thinks before answering
Yes
Open weights
Yes
  • Reasoning
  • Long documents
See the page Nemotron 3 UltraTry Nemotron 3 Ultra

Inkling

Thinking Machines, United StatesOpen model

Model from Thinking Machines, free through OpenRouter. Its provider may train its models on your messages.

Where your message goes

  • OpenRouterFree
    May trainHosted in the US
Context
1,000,000 tokens
Reads images
Yes
Thinks before answering
No
Open weights
Yes
  • Summaries
  • Long documents
See the page InklingTry Inkling

GLM 4.7 Flash

Z.ai (Zhipu), ChinaOpen model

Fast model from Z.ai, served by Cloudflare or, as a fallback, by Z.ai in Singapore.

Where your message goes

  • Cloudflare Workers AIFree
    No trainingGlobal network
  • Z.aiFree
    No trainingHosted in Singapore
Context
Not stated
Reads images
No
Thinks before answering
No
Open weights
Yes
  • Speed
See the page GLM 4.7 FlashTry GLM 4.7 Flash

Providers

The companies that run the models for aidock, with their rules, as published in their official documents.

Technical words are explained further down, in Our method.

Cloudflare Workers AI

Cloudflare, Inc., United States

Training
No training
Message retention
Messages are not kept by default.
Place of processing
Global networkCloudflare global network. No guarantee of processing in the EU.
Data processing agreement
Yes
Transfer outside the EU
EU-US Data Privacy Framework (recertification under review) and standard contractual clauses.
Free quota
10,000 "neurons" a day, shared by all models. Depending on the model, one answer costs 30 to 300.
Terms of use
Production use and end users allowed. Cloudflare does not train models on your content.

Groq

Groq, Inc. (European contract with Groq UK Limited), United States

Training
No training
Message retention
Nothing kept by default. Abuse logs up to 30 days unless zero data retention is enabled.
Place of processing
Hosted in the US
Data processing agreement
Yes
Transfer outside the EU
Standard contractual clauses (Groq is not listed under the Data Privacy Framework).
Free quota
Per model: 30 requests a minute, 1,000 a day, 200,000 tokens a day.
Terms of use
End users allowed, resale forbidden. Preview models are for evaluation only: aidock does not use them.

Mistral AI

Mistral AI, France

Training
May trainThe account holder can opt out. The chat shows live whether aidock has done so.
Message retention
Kept for 30 rolling days to monitor abuse.
Place of processing
Location not guaranteedFrench company. On the free endpoint, Mistral does not guarantee where processing happens.
Data processing agreement
Yes
Transfer outside the EU
EU company. Processing location not guaranteed on the global endpoint.
Free quota
Free mode, no credit card. Limits not published, intended for evaluation.
Terms of use
In free mode, Mistral may train its models on your messages unless the account owner opts out. Preview models train in every case: aidock does not use them.

OpenRouter

OpenRouter, Inc., United States

Training
Depends on the model: each model shows it.
Message retention
OpenRouter keeps nothing without consent. The final provider applies its own rules (shown on each model).
Place of processing
Hosted in the US
Data processing agreement
Not stated
Transfer outside the EU
Standard contractual clauses (OpenRouter is not listed under the Data Privacy Framework).
Free quota
":free" models: 20 requests a minute, 50 a day, or 1,000 a day after a one-time purchase of $10 in credits.
Terms of use
Serving your customers is allowed. Forbidden: reselling access and using several accounts to get around limits.

Scaleway

Scaleway SAS, France

Training
No training
Message retention
Nothing kept. In case of error, content may be kept for up to two weeks.
Place of processing
Hosted in France
Data processing agreement
Yes
Transfer outside the EU
No transfer outside the EU.
Free quota
1,000,000 tokens offered once, then paid (budget capped by aidock).
Terms of use
No restriction found for production use.

Z.ai

Z.ai (Zhipu), China

Training
No training
Message retention
Not stated by the provider.
Place of processing
Hosted in Singapore
Data processing agreement
Not stated
Transfer outside the EU
Transfer outside the EU to Singapore, without an adequacy decision. Contractual safeguards to be checked.
Free quota
"Flash" models are free. Limits not verified.
Terms of use
Your content is not used for training without explicit agreement.

Models we left out and why

Other free offers exist. Here are the ones aidock does not use, and the reason for each.

Our method

  • Catalogue checked on 11 October 2026 in each provider’s official documents (links above).
  • Route: the provider that runs a model for aidock. A model can have several. The chat tries the main route, then switches if it is busy, and tells you.
  • Free: the provider’s free offer. aidock budget: a route paid for by aidock, open only when the budget allows.
  • No training: the provider commits not to train its models on your messages. May train: it may do so, and the chat warns you before your first message.
  • Context: how much text a model reads at once, counted in tokens (a token is about 4 letters).
  • Open weights: the model is published, anyone can run it on their own servers.
  • Endpoint: the technical address where aidock sends your message to the provider. A provider can have several, in different places.
  • Data Privacy Framework: the agreement between the European Union and the United States that allows personal data to be sent to the American companies on its list.
  • Standard contractual clauses: a model contract from the European Commission, signed with the provider to protect data that leaves the European Union.
  • Adequacy decision: the decision by which the European Union recognises that a country protects personal data as well as it does.

For the quality of answers, see compar:IA (new tab), the French government’s public comparison arena, and our Barometer.

Learn to choose the right model

A work situation, an exercise with a real AI model, about 5 minutes.

The right model is the one that does your task.

Try several in the chat, or compare two of them blind.

Try in the chatStart a duel