Compare Models
-
Google
code completion (code-gecko)
$0.002Based on Google’s PaLM 2 large language model, the company specifically trained Codey APIs to handle coding-related prompts, but it also trained the model to handle queries related to Google Cloud. The code completion API provides code autocompletion suggestions as you write code. The API uses the context of the code you’re writing to make its suggestions.
The code completion API supports the code-gecko model. Use the code-gecko model to help improve the speed and accuracy of writing code. The Codey APIs support a wide range of programming languages including C++, C#, Go, GoogleSQL, Java, JavaScript, Kotlin, PHP, Python, Ruby, Rust, Scala, Swift, and TypeScript. You can run with the API and in Generative AI Studio. Some common use cases for code completion include writing code faster, where the code-gecko model is employed to expedite the coding process by leveraging suggested code; and minimizing bugs in code, by utilizing code suggestions that are known to be syntactically correct to circumvent errors, thus reducing the risk of inadvertently introducing bugs that can arise during code creation.
Note: We have converted characters to tokens for the prices (based on the approximation of 4 characters per 1 token).
-
Google
code generation (code-bison)
$0.002Based on Google’s PaLM 2 large language model, the company specifically trained Codey APIs to handle coding-related prompts, but it also trained the model to handle queries related to Google Cloud.
code generation (code-bison) generates code based on a natural language description of the desired code. For example, it can generate a unit test for a function. The code generation API supports the code-bison model. The Codey APIs support a wide range of programming languages, including C++, C#, Go, GoogleSQL, Java, JavaScript, Kotlin, PHP, Python, Ruby, Rust, Scala, Swift, and TypeScript. You can run with the API and in Generative AI Studio.
Some common use cases for code generation include creating unit tests, where you can design a prompt to request a unit test for a specific function; writing a function, which involves passing a problem to the model and receiving a function that solves the problem; and creating a class, where you can use a prompt to describe the purpose of a class and have the code defining that class returned to you.
Note: We have converted characters to tokens for the prices (based on the approximation of 4 characters per 1 token).
-
OpenAI
Curie (fine tuning) GPT-3
$0.012When fine-tuning a GPT model like Curie, you are fine-tuning the GPT-3 base model (not the instruction-oriented variant of GPT-3). Fine-tuning involves taking the pre-trained base model and further training it on your specific dataset or task to enhance its performance. Fine-tuning allows OpenAI API customers to leverage the power of pre-trained GPT-3 language models, such as Curie, while tailoring them to their specific needs (the fine-tuning process allows a model to specialize in a specific task or context, making it more efficient and effective for a particular use case, which can help to reduce costs and latency for high-volume tasks). You are also able to continue fine-tuning a fine-tuned model to add additional data without having to start from scratch.Curie is a larger variant of GPT-3, offering more sophisticated language capabilities. It is a good choice for tasks requiring a deeper understanding of context or more complex language generation. Note: There are two fine-tuning costs to be aware of, a one-time training cost and a pay-as-you-go usage cost. -
OpenAI
Curie Instruct model
$0.002Open AI’s Instruct model Curie is very capable and is faster and costs less than Davinci. Curie can understand and generate natural language. InstructGPT models are sibling models to ChatGPT. They are built on GPT-3 models but made to be safer, more helpful, and more aligned to users’ needs using a technique called reinforcement learning from human feedback (RLHF). Instruct models are meant to generate text with a clear instruction, and they are not optimized for conversational chat. Instruct models are optimized to follow single-turn instructions (e.g., specifically designed to follow instructions provided in a prompt). Developers can use Instruct models for extracting knowledge, generating text, performing NLP tasks, automating tasks involving natural language, and translating languages. Instruct model also make up facts less often than GPT-3 base models and show slight decreases in toxic output generation. Access is available through a request to OpenAI’s API.
-
OpenAI
DALL·E 2
$0.016DALL-E 2 is a browser-based AI system that can create realistic images and art from a description in natural language. It currently supports the ability, given a prompt, to create a new image with a certain size, edit an existing image, or create variations of a user-provided image. Currently, DALL·E 2 charges for an image by pixel resolution.Also to note, for developers, there is also an API available for the beta version and the API allows you to integrate state of the art image generation capabilities directly into your product. The API usage is offered on a pay-as-you-go basis and is billed separately. To note, OpenAI offers large volume discounts (>$5k/month) through their sale team. -
OpenAI
Davinci (fine tuning) GPT-3
$0.12When fine-tuning a GPT model like Davinci, you are fine-tuning the GPT-3 base model (not the instruction-oriented variant of GPT-3). Fine-tuning involves taking the pre-trained base model and further training it on your specific dataset or task to enhance its performance. Fine-tuning allows OpenAI API customers to leverage the power of pre-trained GPT-3 language models, such as Davinci, while tailoring them to their specific needs (the fine-tuning process allows a model to specialize in a specific task or context, making it more efficient and effective for a particular use case, which can help to reduce costs and latency for high-volume tasks). You are also able to continue fine-tuning a fine-tuned model to add additional data without having to start from scratch.Davinci is the largest and most powerful variant of GPT-3. It’s the best choice for tasks requiring the most sophisticated language capabilities, but it also requires more processing power and time to generate results. Note: There are two fine-tuning costs to be aware of, a one-time training cost and a pay-as-you-go usage cost. -
OpenAI
Davinci Instruct model
$0.02Davinci is the most capable Instruct model and it can do any task the other models can (Ada, Babbage and Curie), often with higher quality. InstructGPT models are sibling models to the ChatGPT. They are built on GPT-3 models but made to be safer, more helpful, and more aligned to users’ needs using a technique called reinforcement learning from human feedback (RLHF). Instruct models are meant to generate text with a clear instruction, and they are not optimized for conversational chat. Instruct models are optimized to follow single-turn instructions (e.g., specifically designed to follow instructions provided in a prompt). Developers can use Instruct models for extracting knowledge, generating text, performing NLP tasks, automating tasks involving natural language, and translating languages. Instruct models make up facts less often than GPT-3 base models and show slight decreases in toxic output generation. Access is available through a request to OpenAI’s API. -
Databricks
Dolly 2.0
FREEDolly 2.0 by Databricks, is the first open source, instruction-following Large Language Model, fine-tuned on a human-generated instruction dataset and is licensed for research and commercial use, which means any organization can create, own, and customize powerful LLMs that can talk to people without paying for API access or sharing data with third parties.Dolly 2.0 is a 12B parameter language model based on the EleutherAI pythia model family and fine-tuned exclusively on a new, high-quality human generated instruction following dataset (crowdsourced among Databricks employees – so cool). Dolly-v2-12b is not a state-of-the-art model, but it does exhibit surprisingly high-quality instruction following behavior not characteristic of the foundation model on which it is based. Dolly v2 is also available in smaller model sizes: dolly-v2-7b, a 6.9 billion parameter based on pythia-6.9b and dolly-v2-3b, a 2.8 billion parameter based on pythia-2.8b.Dolly 2.0 can be used for brainstorming, classification, open Q&A, closed Q&A, content generation, information extraction, and summarization. You can access the Dolly 2.0 can training code, the dataset, and the model weights on Hugging Face. -
Google, Stanford University
Electra
FREEELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately) is a transformer-based model like BERT, but it uses a different pre-training approach, which is more efficient and requires less computational resources. It was created by a team of researchers from Google Research, Brain Team, and Stanford University. ELECTRA models are trained to distinguish “real” input tokens vs “fake” input tokens generated by another neural network (for the more technical audience, ELECTRA uses a new pre-training task, called replaced token detection (RTD), that trains a bidirectional model while learning from all input positions). Inspired by generative adversarial networks (GANs), ELECTRA trains the model to distinguish between “real” and “fake” input data. At small scale, ELECTRA achieves strong results even when trained on a single GPU. At large scale, ELECTRA achieves state-of-the-art results on the SQuAD 2.0 dataset. Go to GitHub where you can access the three models (ELECTRA-Small, ELECTRA-Base and ELECTRA-Large). -
Google
FLAN-T5
FREEIf you already know T5, FLAN-T5 is just better at everything. For the same number of parameters, these models have been fine-tuned on more than 1,000 additional tasks covering more languages – the NLP is for English, German, French. It has Apache-2.0 license which is a permissive open source license that allows for commercial use. With appropriate prompting, it can perform zero-shot NLP tasks such as text summarization, common sense reasoning, natural language inference, question answering, sentence and sentiment classification, translation, and pronoun resolution. -
Google
Flan-UL2
FREEDeveloped by Google, Flan-UL2, which is a more powerful version of the T5 model that has been trained using Flan, and it is downloadable from Hugging Face. It shows performance exceeding the ‘prior’ versions of Flan-T5. With the ability to reason for itself and generalize better than the previous models, Flan-UL2 is a great improvement. Flan-UL2 is a machine learning model that can generate textual descriptions of images and has the potential to be used for image search, video captioning, automated content generation, and visual question answering. Flan-UL2 has an Apache-2.0 license, which is a permissive open source license that allows for commercial use.If Flan-UL2’s 20B parameters are too much, consider the previous iteration of Flan-T5, which comes in five different sizes and might be more suitable for your needs. -
Cohere
Generate
$0.015Cohere is a Canadian startup that provides high-performance and secure LLMs for the enterprise. Their models work on public, private, or hybrid clouds.Cohere Generate can be used for tasks such as copywriting, named entity recognition, paraphrasing, and summarization. It can be particularly useful for automating time-consuming and repetitive copywriting tasks and re-wording text to suit a specific reader or context.Cohere Generate is available as an API that can be integrated into various libraries using Python, Node, or Go software development kits (SDKs).We have shown the price of the Cohere Generate Default version, but a Cohere Generate Custom model is available but is double the price (0.030 per 1/k tokens). However, custom models can lead to some of the best-performing NLP models for many tasks.
