Models bump 2025-10

NEWS.md

@@ -1,5 +1,7 @@
 # ellmer (development version)
 
+* `chat_openai()` (and all other providers that use a ChatGPT model) now default to `gpt-5` (#800).
+* `chat_anthropic()` now defaults to `claude-sonnet-4-5-20250929` (#800).
 * New `chat_openai_responses()` to use the new OpenAI responses API (#365).
 * `parallel_chat_structured()` now returns a tibble, since this does a better job of printing more complex data frames (#787).
 * `parallel_chat()` and friends now have a more permissive attitude to errors. By default, they will now return when hitting the first error (rather than erroring), and you can control this behaviour with the `on_error` argument. Or if you interrupt the job, it will finish up current requests and then return all the work done so far. The main downside of this work is that the output of `parallel_chat()` is more complex: it is now a mix of `Chat` objects, error objects, and `NULL` (#628).

R/batch-chat.R

@@ -48,7 +48,7 @@
 #' For any of the aboves, will return `NULL` if `wait = FALSE` and the job
 #' is not complete.
 #' @examplesIf has_credentials("openai")
-#' chat <- chat_openai(model = "gpt-4.1-nano")
+#' chat <- chat_openai(model = "gpt-5-nano")
 #'
 #' # Chat ----------------------------------------------------------------------
 #'

R/provider-anthropic.R

@@ -15,7 +15,7 @@ NULL
 #'
 #' @inheritParams chat_openai
 #' @inherit chat_openai return
-#' @param model `r param_model("claude-sonnet-4-20250514", "anthropic")`
+#' @param model `r param_model("claude-sonnet-4-5-20250929", "anthropic")`
 #' @param api_key `r api_key_param("ANTHROPIC_API_KEY")`
 #' @param max_tokens Maximum number of tokens to generate before stopping.
 #' @param beta_headers Optionally, a character vector of beta headers to opt-in
@@ -43,7 +43,7 @@ chat_anthropic <- function(
 ) {
   echo <- check_echo(echo)
 
-  model <- set_default(model, "claude-sonnet-4-20250514")
+  model <- set_default(model, "claude-sonnet-4-5-20250929")
 
   params <- params %||% params()
   if (lifecycle::is_present(max_tokens)) {
@@ -75,7 +75,7 @@ chat_claude <- chat_anthropic
 
 chat_anthropic_test <- function(
   ...,
-  model = "claude-3-5-sonnet-latest",
+  model = "claude-haiku-4-5-20251001",
   params = NULL,
   echo = "none"
 ) {

R/provider-azure.R

@@ -44,7 +44,7 @@ NULL
 #' @export
 #' @examples
 #' \dontrun{
-#' chat <- chat_azure_openai(model = "gpt-4o-mini")
+#' chat <- chat_azure_openai(model = "gpt-5-nano")
 #' chat$chat("Tell me three jokes about statisticians")
 #' }
 chat_azure_openai <- function(
@@ -128,7 +128,7 @@ chat_azure_openai_test <- function(
     system_prompt = system_prompt,
     api_key = api_key,
     endpoint = "https://ai-hwickhamai260967855527.openai.azure.com",
-    model = "gpt-4o-mini",
+    model = "gpt-5-nano",
     params = params,
     echo = echo
   )

R/provider-github.R

@@ -14,7 +14,7 @@
 #'
 #' @family chatbots
 #' @param api_key `r api_key_param("GITHUB_PAT")`
-#' @param model `r param_model("gpt-4o")`
+#' @param model `r param_model("gpt-5")`
 #' @param params Common model parameters, usually created by [params()].
 #' @export
 #' @inheritParams chat_openai
@@ -37,7 +37,7 @@ chat_github <- function(
 ) {
   check_installed("gitcreds")
 
-  model <- set_default(model, "gpt-4.1")
+  model <- set_default(model, "gpt-5")
   echo <- check_echo(echo)
 
   # https://docs.github.com/en/rest/models/inference?apiVersion=2022-11-28

R/provider-openai-responses.R

@@ -39,7 +39,7 @@ chat_openai_responses <- function(
   api_headers = character(),
   echo = c("none", "output", "all")
 ) {
-  model <- set_default(model, "gpt-4.1")
+  model <- set_default(model, "gpt-5")
   echo <- check_echo(echo)
 
   provider <- ProviderOpenAIResponses(
@@ -56,12 +56,11 @@ chat_openai_responses <- function(
 chat_openai_responses_test <- function(
   system_prompt = "Be terse.",
   ...,
-  model = "gpt-4.1-nano",
+  model = "gpt-5-nano",
   params = NULL,
   echo = "none"
 ) {
   params <- params %||% params()
-  params$temperature <- params$temperature %||% 0
 
   chat_openai_responses(
     system_prompt = system_prompt,

R/provider-openai.R

@@ -16,7 +16,7 @@ NULL
 #' @param system_prompt A system prompt to set the behavior of the assistant.
 #' @param base_url The base URL to the endpoint; the default uses OpenAI.
 #' @param api_key `r api_key_param("OPENAI_API_KEY")`
-#' @param model `r param_model("gpt-4.1", "openai")`
+#' @param model `r param_model("gpt-5", "openai")`
 #' @param params Common model parameters, usually created by [params()].
 #' @param seed Optional integer seed that ChatGPT uses to try and make output
 #'   more reproducible.
@@ -56,7 +56,7 @@ chat_openai <- function(
   api_headers = character(),
   echo = c("none", "output", "all")
 ) {
-  model <- set_default(model, "gpt-4.1")
+  model <- set_default(model, "gpt-5")
   echo <- check_echo(echo)
 
   params <- params %||% params()
@@ -83,13 +83,12 @@ chat_openai <- function(
 chat_openai_test <- function(
   system_prompt = "Be terse.",
   ...,
-  model = "gpt-4.1-nano",
+  model = "gpt-5-nano",
   params = NULL,
   echo = "none"
 ) {
   params <- params %||% params()
   params$seed <- params$seed %||% 1014
-  params$temperature <- params$temperature %||% 0
 
   chat_openai(
     system_prompt = system_prompt,

R/provider-openrouter.R

@@ -12,7 +12,7 @@ NULL
 #' @export
 #' @family chatbots
 #' @param api_key `r api_key_param("OPENROUTER_API_KEY")`
-#' @param model `r param_model("gpt-4o")`
+#' @param model `r param_model("gpt-5")`
 #' @param params Common model parameters, usually created by [params()].
 #' @inheritParams chat_openai
 #' @inherit chat_openai return
@@ -31,7 +31,7 @@ chat_openrouter <- function(
   echo = c("none", "output", "all"),
   api_headers = character()
 ) {
-  model <- set_default(model, "gpt-4o")
+  model <- set_default(model, "gpt-5")
   echo <- check_echo(echo)
 
   params <- params %||% params()

R/provider-portkey.R

@@ -11,7 +11,7 @@
 #'
 #' @family chatbots
 #' @param api_key `r api_key_param("PORTKEY_API_KEY")`
-#' @param model `r param_model("gpt-4o", "openai")`
+#' @param model `r param_model("gpt-5", "openai")`
 #' @param virtual_key A virtual identifier storing LLM provider's API key. See
 #'   [documentation](https://portkey.ai/docs/product/ai-gateway/virtual-keys).
 #'   Can be read from the `PORTKEY_VIRTUAL_KEY` environment variable.
@@ -34,7 +34,7 @@ chat_portkey <- function(
   echo = NULL,
   api_headers = character()
 ) {
-  model <- set_default(model, "gpt-4o")
+  model <- set_default(model, "gpt-5")
   echo <- check_echo(echo)
 
   params <- params %||% params()
@@ -53,7 +53,7 @@ chat_portkey <- function(
 
 chat_portkey_test <- function(
   ...,
-  model = "gpt-4o-mini",
+  model = "gpt-5-nano",
   params = NULL,
   echo = "none"
 ) {

README.Rmd

@@ -67,7 +67,7 @@ If you're using ellmer inside an organisation, you may have internal policies th
 
 If you're using ellmer for your own exploration, you'll have a lot more freedom, so we have a few recommendations to help you get started:
 
-- `chat_openai()` or `chat_anthropic()` are good places to start. `chat_openai()` defaults to **GPT-4.1**, but you can use `model = "gpt-4-1-nano"` for a cheaper, faster model, or `model = "o3"` for more complex reasoning. `chat_anthropic()` is also good; it defaults to **Claude 4.0 Sonnet**, which we have found to be particularly good at writing R code.
+- `chat_openai()` or `chat_anthropic()` are good places to start. `chat_openai()` defaults to **GPT-5**, but you can use `model = "gpt-5-mini"` or `model = "gpt-5-nano"` for a cheaper, faster models. `chat_anthropic()` is also good; it defaults to **Claude 4.5 Sonnet**, which we have found to be particularly good at writing R code.
 
 - `chat_google_gemini()` is a strong model with generous free tier (with the downside that [your data is used](https://ai.google.dev/gemini-api/terms#unpaid-services) to improve the model), making it a great place to start if you don't want to spend any money.
 
@@ -88,7 +88,7 @@ You can work with ellmer in several different ways, depending on whether you are
 ```{r}
 library(ellmer)
 
-chat <- chat_openai("Be terse", model = "gpt-4o-mini")
+chat <- chat_openai("Be terse", model = "gpt-5-mini")
 ```
 
 Chat objects are stateful [R6 objects](https://r6.r-lib.org): they retain the context of the conversation, so each new query builds on the previous ones. You call their methods with `$`.
@@ -141,7 +141,7 @@ In most circumstances, ellmer will stream the output to the console. You can tak
 
 ```{r}
 my_function <- function() {
-  chat <- chat_openai("Be terse", model = "gpt-4o-mini", echo = "none")
+  chat <- chat_openai("Be terse", model = "gpt-5-mini", echo = "none")
   chat$chat("What is 6 times 7?")
 }
 str(my_function())