​

Selecting a model

• Whenever you start a new chat, you can use the model you want to work with at the top left.
• You can still change the model at the top left if you have already started a chat. For example, you can start with GPT-4o and, after three messages, switch to Claude 3 Sonnnet.
• When you switch models in an ongoing chat, the entire context of the previous chat history and your data (documents, texts, websites) is always passed to the selected model. So, you don’t need to worry about keeping the context when switching models.
• As no data is ever stored in the models, the system provides all context to the model with every request you make.
• You can also set your personal default model in the account settings here. The default for new users is GPT-4o.

​

Recommendations

Generally, we recommend our users to use GPT-4o. This is the strongest all-round model right now and provides good results in most use cases. As an alternative, you can also try CLaude 3.5 Sonnet. This model is preferred over GPT-4o by many users for software engineering tasks and text creation (emails, descriptions, documentation,…).

For using the data analyst functionality, we recommend GPT-4o.

When working with very large files (>128k tokens), we recommend to use Gemini 1.5 Pro, as this model has the largest context window, i.e. can fit the longest texts to be processed.

The other models are good as well in translating, answering smaller questions or for specific questions. Feel free to try around to get a feeling for the differences - for example between the European Mistral models to the most famous GPT-models or the open-source models of Meta - and find the best model for your use case!

​

Model overview

Here is a description of the most-used models:

​

OpenAI

OpenAI is the most famous AI company and known for its advanced language models, the GPT series. OpenAI also emphasizes ethical AI development, ensuring that their technologies are used responsibly and align with human values.

​

o1-preview

The new omni-models are designed to think before answering a user query. This enables them to reason through complex reasoning tasks and makes them better at coding, math and science tasks. Before answering, they will spend time thinking about the task. Because of this, they are not designed to replace the previous GPT-4o models, but work as a complementary model for specific tasks.

In the first version, they can not process files, use custom instructions, perform data analyses, search the web or analyse images (vision capability). They have a context window of 128k tokens. o1-preview has an output limit of around 65k tokens.

​

o1-mini

o1-mini is the smaller, faster model. It still has good reasoning capabilities, although a bit lower than o1-preview due to the smaller size. The context window is 128k tokens as well and the output limit is around 32k tokens.

​

GPT-4o

GPT-4o is the latest and best model from OpenAI. It has a similar quality as the previous version of GPT-4 (in rare cases a slightly worse quality was observed), but operates at a higher speed - similar to GPT-3.5. It is the most used model on Langdock and a good allrounder with satisfactory performance in most situations.

GPT-4o is capable of analyzing, describing or comparing images. Simply press the upload button in the chat input field or drag and drop an image into the field to upload it.

You can also set the model as your personal default model in your preferences. This will automatically pre-select the model in any new conversation.

​

GPT-4o mini

This is the smaller version of GPT-4o, performing at a similar quality level. The key difference is the higher speed - it’s significantly faster compared to GPT-4o, but due to the smaller size less capable in complex tasks or reasoning.

​

GPT-4

The older version of GPT-4 was released in early 2024. It still provides qualitative outputs but at a slower speed than GPT-4o. There are exceptions, where GPT-4 provides more suitable results than the newer version, but for most users we recommend switching to GPT-4o.

GPT-4 is not able to analyze images. Please switch to GPT-4o or the Claude models for that.

​

GPT-3.5

The older model from OpenAI is GPT-3.5. When it came out, it was the best model and fascinated users with its intelligence and capabilities. By now, the newer models from OpenAI, Anthropic, and Mistral have a higher quality, with sometimes an equal or even better speed. GPT-3.5 will be deprecated eventually.

​

DALL-E 3

For generating images, we use the latest image generation model from OpenAI. All models (including the ones not from OpenAI) are able to call the model and have an image generated. Each individual model writes a prompt based on your that is sent to the image model. As these prompts might differ from model to model, it makes sense to try different models. The response also includes a description of the image, which is not generated by DALL-E, but the other models.

We recommend GPT-4o or Claude 3.5 Sonnet for the best results based on our experience and feedback from users.

​

Anthropic

Anthropic is an AI research company focused on creating AI systems that are interpretable, reliable, and aligned with human intentions. Their models are designed with a strong emphasis on safety and robustness, aiming to address risks associated with AI technologies.

​

Claude 3

The Claude 3 family is built by Anthropic and according to some sources more trustworthy and reliable than other models. It hallucinates less and is better at some reasoning tasks. The family consists of three different models, each with their performance levels (speed, quality and costs), which allows users to choose the right model for their specific need.

Generally, the models are good in these areas and, in some areas, even better than GPT-4:

• Text creation: The tone and style are more natural and human than texts from GPT-4.
• Refusing to answer: GPT-4 sometimes refuses to answer and says it could not access the web / the attached document / analyze the provided image. We have not observed this with Claude.
• Image analysis/vision: The Claude models can also analyze uploaded images.
• Coding: Claude 3 seems to be better in some reasoning and coding tasks.

​

Haiku

The smallest model of the family is comparable to GPT-3.5 in its quality. It might reach its limits with reasoning or more nuanced tasks, but is good for translations, instant answers and extraction of knowledge from data.

​

Sonnet

The second model, Sonnet, balances intelligence and speed which makes it the go-to model for most cases. We recommend it especially for text creation and every use case where you need some written output as the answers sound more human than from other AI models.

​

Opus

The larger Opus model is good at performing complex reasoning tasks, like research, drug discovery or coding. It also performs well in strategic analyses and analysing graphs and trends.

​

Claude 3.5 Sonnet

This new Claude model operates at the same speed as Claude Sonnet but offers significantly higher quality. It outperforms Anthropic’s previous top model, Opus, and even surpasses GPT-4o in some areas. It is recommended for complex tasks like coding, strategy, reasoning, and text generation.

It is also considered to be the first equally good model to the OpenAI models.

​

Gemini

Gemini models are AI language models developed by Google DeepMind. They are noted for their capabilities in generating human-like text, understanding complex queries and processing large amounts of data. They have a very large context window (=the amount of text the model can process at once) of 1 million or 2 million tokens. This is impressive compared to the other models, most of which have a context limit of 128k tokens.

​

Gemini 1.5 Pro

Gemini 1.5 Pro is designed for complex tasks with reasoning and analysis. It supports tool calling and is suitable for complex tasks like reasoning or coding. This model has an even higher context window of 2 million tokens. For comparison: Most of the other models available right now (including GPT-4o) have a context window of 128 thousand tokens. This makes it suitable to work with large documents that previously did not fit into the context window of other models.

​

Gemini 1.5 Flash

Gemini 1.5 Flash is a fast model with rapid processing which excels in summarization, chat and data extraction. The downside of the speed is that it is not as strong in tool calling (document search, web search, data analyst etc.) - same as the other smaller models (Claude Haiku, GPT-4o mini etc.). Instead, it is suitable for summarization, classification or translations. The model still has an impressive context window (=the amount of text the model can process at once) of 1 Million tokens.

​

Mistral

Mistral is a French AI company focused on developing language models that excel in both performance and efficiency. The company was praised for its open-source approach with many models, increasing transparency for users and researchers. The latest Mistral models also have a 128k token context window. Especially the Mistral Large 2 model is an improvement to Mistral’s previous Mistral Large model:

​

Mistral Large 2

The model significantly improved compared to the previous Mistral Large, especially in code generation, mathematics reasoning, native multi-lingual support and tool-calling. It has good reasoning capabilities, supports dozens of languages and 80+ coding languages. The model performs similarly to the other flagship models, GPT-4o, Claude 3.5 and Llama 3.1 405B, especially in coding and math tasks.

​

Mistral NeMo

The NeMo model was built in collaboration with NVIDIA and has 12B parameters, making it smaller than the Mistral Large 2 model. It has sound reasoning and coding capabilities because of its size, making it suitable for translation and summarization

​

Llama

The Llama models are built by Meta and the best open-source modelsThe Llama 3.1 models are a significant moment in AI development. They are open-source AI models (meaning the source code is publicly accessible and reviewable), surpassing most other open-source and even some closed-source models. They all have a 128k token context window (meaning they can process the same amount of text or documents as GPT-4o) and excel in language tasks, contextual understanding and multi-step tasks.

​

Llama 3.1 8B

The 8B model is suitable for lightweight use cases like classification or translations as it supports eight languages (including English, German, French, and Spanish).

​

Llama 3.1 70B

The 70B model is ideal for more nuanced tasks like content writing, text summarization, following instructions (from your assistants), writing computer code or solving math problems.

​

Llama 3.1 405B (coming soon)

We will add the Llama 3.1 405B model (rivalling the top models GPT-4o and Claude Sonnet) in the next weeks, as soon as it is available in the EU!

​

Context Window Sizes

Model	Context Window Size
GPT-4o	128k tokens
GPT-4o mini	128k tokens
GPT-4 Turbo	128k tokens
o1-previw	128k tokens
o1-mini	128k tokens
GPT-3.5	16k tokens
Claude 3.5 Sonnet	200k tokens
Claude 3 Haiku	200k tokens
Claude 3 Sonnet	200k tokens
Claude 3 Opus	200k tokens
Gemini 1.5 Pro	2m tokens
Gemini 1.5 Flash	1m tokens
Mistral Large 2	128k tokens
Mistral NeMo	128k tokens
Llama 3.1 8B	128k tokens
Llama 3.1 70B	128k tokens