> This page is for version v2 API (default). > For other versions, use one of these documentation indexes: > - v2 API (default): https://docs.cohere.com/v2/llms.txt > - v1 API: https://docs.cohere.com/v1/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.cohere.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cohere.com/_mcp/server. # Generate POST https://api.cohere.com/v1/generate Content-Type: application/json This API is marked as "Legacy" and is no longer maintained. Follow the [migration guide](https://docs.cohere.com/docs/migrating-from-cogenerate-to-cochat) to start using the Chat API. Generates realistic text conditioned on a given input. Reference: https://docs.cohere.com/reference/generate-v1 ## Authentication - `Authorization` header (bearer token, required) — Bearer authentication of the form `Bearer `, where token is your auth token. ## Request ### Headers - `X-Client-Name` (string, optional) — The name of the project that is making the request. ### Body (application/json) This endpoint expects an object. - `prompt` (string, required) — The input text that serves as the starting point for generating the response. Note: The prompt will be pre-processed and modified before reaching the model. - `stream` (false, required) — When `true`, the response will be a JSON stream of events. Streaming is beneficial for user interfaces that render the contents of the response piece by piece, as it gets generated. The final event will contain the complete response, and will contain an `is_finished` field set to `true`. The event will also contain a `finish_reason`, which can be one of the following: - `COMPLETE` - the model sent back a finished reply - `MAX_TOKENS` - the reply was cut off because the model reached the maximum number of tokens for its context length - `ERROR` - something went wrong when generating the reply - `ERROR_TOXIC` - the model generated a reply that was deemed toxic - `model` (string, optional) — The identifier of the model to generate with. Currently available models are `command` (default), `command-nightly` (experimental), `command-light`, and `command-light-nightly` (experimental). Smaller, "light" models are faster, while larger models will perform better. [Custom models](https://docs.cohere.com/docs/training-custom-models) can also be supplied with their full ID. - `num_generations` (integer, optional) — The maximum number of generations that will be returned. Defaults to `1`, min value of `1`, max value of `5`. - `max_tokens` (integer, optional) — The maximum number of tokens the model will generate as part of the response. Note: Setting a low value may result in incomplete generations. This parameter is off by default, and if it's not specified, the model will continue generating until it emits an EOS completion token. See [BPE Tokens](/bpe-tokens-wiki) for more details. Can only be set to `0` if `return_likelihoods` is set to `ALL` to get the likelihood of the prompt. - `truncate` (enum, optional, default: END) — One of `NONE|START|END` to specify how the API will handle inputs longer than the maximum token length. Passing `START` will discard the start of the input. `END` will discard the end of the input. In both cases, input is discarded until the remaining input is exactly the maximum input token length for the model. If `NONE` is selected, when the input exceeds the maximum input token length an error will be returned. - Allowed values: `NONE`, `START`, `END` - `temperature` (double, optional) — A non-negative float that tunes the degree of randomness in generation. Lower temperatures mean less random generations. See [Temperature](/temperature-wiki) for more details. Defaults to `0.75`, min value of `0.0`, max value of `5.0`. - `seed` (integer, optional) — If specified, the backend will make a best effort to sample tokens deterministically, such that repeated requests with the same seed and parameters should return the same result. However, determinism cannot be totally guaranteed. Compatible Deployments: Cohere Platform, Azure, AWS Sagemaker/Bedrock, Private Deployments - `preset` (string, optional) — Identifier of a custom preset. A preset is a combination of parameters, such as prompt, temperature etc. You can create presets in the [playground](https://dashboard.cohere.com/playground/generate). When a preset is specified, the `prompt` parameter becomes optional, and any included parameters will override the preset's parameters. - `end_sequences` (list of string, optional) — The generated text will be cut at the beginning of the earliest occurrence of an end sequence. The sequence will be excluded from the text. - `stop_sequences` (list of string, optional) — The generated text will be cut at the end of the earliest occurrence of a stop sequence. The sequence will be included the text. - `k` (integer, optional) — Ensures only the top `k` most likely tokens are considered for generation at each step. Defaults to `0`, min value of `0`, max value of `500`. - `p` (double, optional) — Ensures that only the most likely tokens, with total probability mass of `p`, are considered for generation at each step. If both `k` and `p` are enabled, `p` acts after `k`. Defaults to `0.75`. min value of `0.01`, max value of `0.99`. - `frequency_penalty` (double, optional) — Used to reduce repetitiveness of generated tokens. The higher the value, the stronger a penalty is applied to previously present tokens, proportional to how many times they have already appeared in the prompt or prior generation. Using `frequency_penalty` in combination with `presence_penalty` is not supported on newer models. - `presence_penalty` (double, optional) — Defaults to `0.0`, min value of `0.0`, max value of `1.0`. Can be used to reduce repetitiveness of generated tokens. Similar to `frequency_penalty`, except that this penalty is applied equally to all tokens that have already appeared, regardless of their exact frequencies. Using `frequency_penalty` in combination with `presence_penalty` is not supported on newer models. - `return_likelihoods` (enum, optional, default: NONE) — One of `GENERATION|NONE` to specify how and if the token likelihoods are returned with the response. Defaults to `NONE`. If `GENERATION` is selected, the token likelihoods will only be provided for generated text. WARNING: `ALL` is deprecated, and will be removed in a future release. - Allowed values: `GENERATION`, `ALL`, `NONE` - `raw_prompting` (boolean, optional) — When enabled, the user's prompt will be sent to the model without any pre-processing. ## Response ### 200 - `id` (string, required) - `generations` (list of SingleGeneration, required) — List of generated results - `prompt` (string, optional) — Prompt used for generations. - `meta` (ApiMeta, optional) ## Errors ### 400 Bad Request Error This error is returned when the request is not well formed. This could be because: - JSON is invalid - The request is missing required fields - The request contains an invalid combination of fields - `message` (string, optional) - `id` (string, optional) ### 401 Unauthorized Error This error indicates that the operation attempted to be performed is not allowed. This could be because: - The api token is invalid - The user does not have the necessary permissions - `message` (string, optional) - `id` (string, optional) ### 403 Forbidden Error This error indicates that the operation attempted to be performed is not allowed. This could be because: - The api token is invalid - The user does not have the necessary permissions - `message` (string, optional) - `id` (string, optional) ### 404 Not Found Error This error is returned when a resource is not found. This could be because: - The endpoint does not exist - The resource does not exist eg model id, dataset id - `message` (string, optional) - `id` (string, optional) ### 422 Unprocessable Entity Error This error is returned when the request is not well formed. This could be because: - JSON is invalid - The request is missing required fields - The request contains an invalid combination of fields - `message` (string, optional) - `id` (string, optional) ### 429 Too Many Requests Error Too many requests - `message` (string, optional) - `id` (string, optional) ### 498 Invalid Token Error This error is returned when a request or response contains a deny-listed token. - `message` (string, optional) - `id` (string, optional) ### 499 Client Closed Request Error This error is returned when a request is cancelled by the user. - `message` (string, optional) - `id` (string, optional) ### 500 Internal Server Error This error is returned when an uncategorised internal server error occurs. - `message` (string, optional) - `id` (string, optional) ### 501 Not Implemented Error This error is returned when the requested feature is not implemented. - `message` (string, optional) - `id` (string, optional) ### 503 Service Unavailable Error This error is returned when the service is unavailable. This could be due to: - Too many users trying to access the service at the same time - `message` (string, optional) - `id` (string, optional) ### 504 Gateway Timeout Error This error is returned when a request to the server times out. This could be due to: - An internal services taking too long to respond - `message` (string, optional) - `id` (string, optional) ## Types ### SingleGeneration - `id` (string, required) - `text` (string, required) - `index` (integer, optional) — Refers to the nth generation. Only present when `num_generations` is greater than zero. - `likelihood` (double, optional) - `token_likelihoods` (list of SingleGenerationTokenLikelihoodsItems, optional) — Only returned if `return_likelihoods` is set to `GENERATION` or `ALL`. The likelihood refers to the average log-likelihood of the entire specified string, which is useful for [evaluating the performance of your model](likelihood-eval), especially if you've created a [custom model](https://docs.cohere.com/docs/training-custom-models). Individual token likelihoods provide the log-likelihood of each token. The first token will not have a likelihood. ### ApiMeta - `api_version` (ApiMetaApiVersion, optional) - `billed_units` (ApiMetaBilledUnits, optional) - `tokens` (ApiMetaTokens, optional) - `cached_tokens` (double, optional) — The number of prompt tokens that hit the inference cache. - `warnings` (list of string, optional) ### SingleGenerationTokenLikelihoodsItems - `token` (string, required) - `likelihood` (double, required) ### ApiMetaApiVersion - `version` (string, required) - `is_deprecated` (boolean, optional) - `is_experimental` (boolean, optional) ### ApiMetaBilledUnits - `images` (double, optional) — The number of billed images. - `input_tokens` (double, optional) — The number of billed input tokens. - `image_tokens` (double, optional) — The number of billed image tokens. - `output_tokens` (double, optional) — The number of billed output tokens. - `search_units` (double, optional) — The number of billed search units. - `classifications` (double, optional) — The number of billed classifications units. - `pages` (double, optional) — The number of billed pages parsed. ### ApiMetaTokens - `input_tokens` (double, optional) — The number of tokens used as input to the model. - `output_tokens` (double, optional) — The number of tokens produced by the model. ## Examples **Request** ```json { "prompt": "Please explain to me how LLMs work", "stream": false } ``` **Response** ```json { "id": "6afae9c2-3375-4d0e-8d18-2e9eb7f2c3ec", "generations": [ { "id": "8e6de35d-3007-43ab-9253-ac4f95dcb8a2", "text": "LLMs, or Large Language Models, are a type of neural network-based AI model that has been trained on massive amounts of text data and have become ubiquitous in the AI landscape. They possess astounding capabilities for comprehending and generating human-like language.\nThese models leverage neural networks that operate on a large scale, often involving millions or even billions of parameters. This substantial scale enables them to capture intricate patterns and connections within the vast amounts of text they have been trained on.\n\nThe training process for LLMs is fueled by colossal datasets of textual information, ranging from books and articles to websites and conversational transcripts. This extensive training enables them to develop a nuanced understanding of language patterns, grammar, and semantics.\n\nWhen posed with a new text input, LLMs employ their finely honed understanding of language to generate informed responses or undertake tasks such as language translation, text completion, or question answering. They do this by manipulating the input text through adding, removing, or altering elements to craft a desired output.\n\nOne of the underlying principles of their efficacy is the recurrent neural network (RNN) architecture they often adopt. This design enables them to process sequential data like natural language effectively. RNNs possess \"memory\" aspects via loops between layers, which allows them to retain and manipulate information gathered across long sequences, akin to the way humans process information.\n\nHowever, it's their size that arguably constitutes their most notable aspect. The sheer volume of these models – with counts of parameters often exceeding 100 million – enables them to capture correlations and patterns within language data effectively. This empowers them to generate coherent and contextually appropriate responses, posing a remarkable advancement in conversational AI.\n\nWhile LLMs have demonstrated extraordinary language prowess, it's vital to acknowledge their limitations and potential for improvement. Their biases often reflect those of the training data, and they may struggle with logical inconsistencies or factual errors. Ongoing research aims to enhance their robustness, diversity, and overall usability.\n\nIn essence, LLMs are a groundbreaking manifestation of AI's potential to simulate and even extend human language capabilities, while also serving as a testament to the ongoing journey towards refining and perfecting these technologies." } ], "prompt": "Please explain to me how LLMs work", "meta": { "api_version": { "version": "1" }, "billed_units": { "input_tokens": 8, "output_tokens": 442 } } } ``` **SDK Code** ```go Cohere Go SDK package main import ( "context" "errors" "io" "log" "os" cohere "github.com/cohere-ai/cohere-go/v2" client "github.com/cohere-ai/cohere-go/v2/client" ) func main() { co := client.NewClient(client.WithToken(os.Getenv("CO_API_KEY"))) resp, err := co.GenerateStream( context.TODO(), &cohere.GenerateStreamRequest{ Prompt: "Please explain to me how LLMs work", }, ) if err != nil { log.Fatal(err) } // Make sure to close the stream when you're done reading. // This is easily handled with defer. defer resp.Close() for { message, err := resp.Recv() if errors.Is(err, io.EOF) { // An io.EOF error means the server is done sending messages // and should be treated as a success. break } if message.TextGeneration != nil { log.Printf("%+v", resp) } } } ``` ```typescript Cohere TypeScript SDK import { CohereClient } from 'cohere-ai'; const cohere = new CohereClient({}); (async () => { const generate = await cohere.generate({ prompt: 'Please explain to me how LLMs work', }); console.log(generate); })(); ``` ```python Sync import cohere co = cohere.Client() response = co.generate( prompt="Please explain to me how LLMs work", ) print(response) ``` ```python Async import cohere import asyncio co = cohere.AsyncClient() async def main(): response = await co.generate( prompt="Please explain to me how LLMs work", ) print(response) asyncio.run(main()) ``` ```java Cohere java SDK /* (C)2024 */ import com.cohere.api.Cohere; import com.cohere.api.requests.GenerateRequest; import com.cohere.api.types.Generation; public class GeneratePost { public static void main(String[] args) { Cohere cohere = Cohere.builder().clientName("snippet").build(); Generation response = cohere.generate( GenerateRequest.builder().prompt("Please explain to me how LLMs work").build()); System.out.println(response); } } ``` ```ruby require 'uri' require 'net/http' url = URI("https://api.cohere.com/v1/generate") http = Net::HTTP.new(url.host, url.port) http.use_ssl = true request = Net::HTTP::Post.new(url) request["Authorization"] = 'Bearer ' request["Content-Type"] = 'application/json' request.body = "{\n \"prompt\": \"Please explain to me how LLMs work\",\n \"stream\": false\n}" response = http.request(request) puts response.read_body ``` ```php request('POST', 'https://api.cohere.com/v1/generate', [ 'body' => '{ "prompt": "Please explain to me how LLMs work", "stream": false }', 'headers' => [ 'Authorization' => 'Bearer ', 'Content-Type' => 'application/json', ], ]); echo $response->getBody(); ``` ```csharp using RestSharp; var client = new RestClient("https://api.cohere.com/v1/generate"); var request = new RestRequest(Method.POST); request.AddHeader("Authorization", "Bearer "); request.AddHeader("Content-Type", "application/json"); request.AddParameter("application/json", "{\n \"prompt\": \"Please explain to me how LLMs work\",\n \"stream\": false\n}", ParameterType.RequestBody); IRestResponse response = client.Execute(request); ``` ```swift import Foundation let headers = [ "Authorization": "Bearer ", "Content-Type": "application/json" ] let parameters = [ "prompt": "Please explain to me how LLMs work", "stream": false ] as [String : Any] let postData = JSONSerialization.data(withJSONObject: parameters, options: []) let request = NSMutableURLRequest(url: NSURL(string: "https://api.cohere.com/v1/generate")! as URL, cachePolicy: .useProtocolCachePolicy, timeoutInterval: 10.0) request.httpMethod = "POST" request.allHTTPHeaderFields = headers request.httpBody = postData as Data let session = URLSession.shared let dataTask = session.dataTask(with: request as URLRequest, completionHandler: { (data, response, error) -> Void in if (error != nil) { print(error as Any) } else { let httpResponse = response as? HTTPURLResponse print(httpResponse) } }) dataTask.resume() ```