diff --git a/NEWS.md b/NEWS.md
index 07dba40c..71e933fc 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,5 +1,7 @@
 # ellmer (development version)
 
+* `chat_portkey()` now requires you to supply a model (#786).
+* `chat_portkey(virtual_key)` no longer needs to be supplied; instead Portkey recommends including the virtual key/povider in the `model`.(#786).
 * `batch_chat()` logs tokens once, on retrieval (#743).
 * `params()` gains new `reasoning_effort` and `reasoning_tokens` so you can control the amount of effort a model spends on thinking. Initial support is provided for `chat_claude()`, `chat_google_gemini()`, and `chat_openai()` (#720).
 * `chat_anthropic()` gains new `cache` parameter to control caching. By default it is set to "5m". This should (on average) reduce the cost of your chats.(#584)
diff --git a/R/provider-portkey.R b/R/provider-portkey.R
index e00c73cf..e76b90a9 100644
--- a/R/provider-portkey.R
+++ b/R/provider-portkey.R
@@ -3,42 +3,61 @@
 #' @description
 #' [PortkeyAI](https://portkey.ai/docs/product/ai-gateway/universal-api)
 #' provides an interface (AI Gateway) to connect through its Universal API to a
-#' variety of LLMs providers with a single endpoint.
-#'
-#' ## Authentication
-#' API keys together with configurations of LLM providers are
-#' stored inside Portkey application.
+#' variety of LLMs providers via a single endpoint.
 #'
 #' @family chatbots
+#' @param model The model name, e.g. `@my-provider/my-model`.
 #' @param api_key `r lifecycle::badge("deprecated")` Use `credentials` instead.
 #' @param credentials `r api_key_param("PORTKEY_API_KEY")`
-#' @param model `r param_model("gpt-4o", "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.
+#' @param virtual_key `r lifecycle::badge("deprecated")`.
+#'   Portkey now recommend supplying the model provider
+#'   (formerly known as the `virtual_key`), in the model name, e.g.
+#'   `@my-provider/my-model`. See
+#'   <https://portkey.ai/docs/support/upgrade-to-model-catalog> for details.
+#'
+#'   For backward compatibility, the `PORTKEY_VIRTUAL_KEY` env var is still used
+#'   if the model
+#'  you can specify the virtual key here.
 #' @export
 #' @inheritParams chat_openai
 #' @inherit chat_openai return
 #' @examples
 #' \dontrun{
-#' chat <- chat_portkey(virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"))
+#' chat <- chat_portkey()
 #' chat$chat("Tell me three jokes about statisticians")
 #' }
 chat_portkey <- function(
+  model,
   system_prompt = NULL,
   base_url = "https://api.portkey.ai/v1",
   api_key = NULL,
   credentials = NULL,
-  virtual_key = portkey_virtual_key(),
-  model = NULL,
+  virtual_key = deprecated(),
   params = NULL,
   api_args = list(),
   echo = NULL,
   api_headers = character()
 ) {
-  model <- set_default(model, "gpt-4o")
+  check_string(model)
   echo <- check_echo(echo)
 
+  if (lifecycle::is_present(virtual_key)) {
+    lifecycle::deprecate_warn(
+      when = "0.4.0",
+      what = "chat_portkey(virtual_key=)",
+      with = "chat_portkey(model=)",
+    )
+    check_string(virtual_key, allow_null = TRUE)
+  } else {
+    virtual_key <- NULL
+  }
+
+  # For backward compatibility
+  if (!grepl("^@", model)) {
+    virtual_key <- virtual_key %||% key_get("PORTKEY_VIRTUAL_KEY")
+    model <- paste0("@", virtual_key, "/", model)
+  }
+
   credentials <- as_credentials(
     "chat_portkey",
     function() portkey_key(),
@@ -54,7 +73,6 @@ chat_portkey <- function(
     params = params,
     extra_args = api_args,
     credentials = credentials,
-    virtual_key = virtual_key,
     extra_headers = api_headers
   )
   Chat$new(provider = provider, system_prompt = system_prompt, echo = echo)
@@ -62,7 +80,7 @@ chat_portkey <- function(
 
 chat_portkey_test <- function(
   ...,
-  model = "gpt-4o-mini",
+  model = "@open-ai-virtual-7f0dcd/gpt-4o-mini",
   params = NULL,
   echo = "none"
 ) {
@@ -74,29 +92,13 @@ chat_portkey_test <- function(
 
 ProviderPortkeyAI <- new_class(
   "ProviderPortkeyAI",
-  parent = ProviderOpenAI,
-  properties = list(
-    virtual_key = prop_string(allow_null = TRUE)
-  )
+  parent = ProviderOpenAI
 )
 
-portkey_key_exists <- function() {
-  key_exists("PORTKEY_API_KEY")
-}
-
 portkey_key <- function() {
   key_get("PORTKEY_API_KEY")
 }
 
-portkey_virtual_key <- function() {
-  val <- Sys.getenv("PORTKEY_VIRTUAL_KEY")
-  if (!identical(val, "")) {
-    val
-  } else {
-    NULL
-  }
-}
-
 method(base_request, ProviderPortkeyAI) <- function(provider) {
   req <- request(provider@base_url)
 
@@ -105,7 +107,6 @@ method(base_request, ProviderPortkeyAI) <- function(provider) {
     provider@credentials(),
     "x-portkey-api-key"
   )
-  req <- httr2::req_headers(req, `x-portkey-virtual-key` = provider@virtual_key)
   req <- ellmer_req_robustify(req)
   req <- ellmer_req_user_agent(req)
   req <- base_request_error(provider, req)
@@ -117,15 +118,13 @@ method(base_request, ProviderPortkeyAI) <- function(provider) {
 #' @rdname chat_portkey
 models_portkey <- function(
   base_url = "https://api.portkey.ai/v1",
-  api_key = portkey_key(),
-  virtual_key = NULL
+  api_key = portkey_key()
 ) {
   provider <- ProviderPortkeyAI(
     name = "PortkeyAI",
     model = "",
     base_url = base_url,
-    credentials = function() api_key,
-    virtual_key = virtual_key
+    credentials = function() api_key
   )
 
   req <- base_request(provider)
@@ -135,11 +134,11 @@ models_portkey <- function(
   json <- resp_body_json(resp)
 
   id <- map_chr(json$data, "[[", "id")
-  created_at <- as.POSIXct(map_dbl(json$data, "[[", "created_at"))
+  slug <- map_chr(json$data, "[[", "slug")
 
   df <- data.frame(
     id = id,
-    created_at = created_at
+    slug = slug
   )
-  df[order(-xtfrm(df$created_at)), ]
+  df
 }
diff --git a/man/chat_portkey.Rd b/man/chat_portkey.Rd
index e8d5e0fd..6bd1ad64 100644
--- a/man/chat_portkey.Rd
+++ b/man/chat_portkey.Rd
@@ -6,25 +6,23 @@
 \title{Chat with a model hosted on PortkeyAI}
 \usage{
 chat_portkey(
+  model,
   system_prompt = NULL,
   base_url = "https://api.portkey.ai/v1",
   api_key = NULL,
   credentials = NULL,
-  virtual_key = portkey_virtual_key(),
-  model = NULL,
+  virtual_key = deprecated(),
   params = NULL,
   api_args = list(),
   echo = NULL,
   api_headers = character()
 )
 
-models_portkey(
-  base_url = "https://api.portkey.ai/v1",
-  api_key = portkey_key(),
-  virtual_key = NULL
-)
+models_portkey(base_url = "https://api.portkey.ai/v1", api_key = portkey_key())
 }
 \arguments{
+\item{model}{The model name, e.g. \verb{@my-provider/my-model}.}
+
 \item{system_prompt}{A system prompt to set the behavior of the assistant.}
 
 \item{base_url}{The base URL to the endpoint; the default uses OpenAI.}
@@ -37,13 +35,15 @@ You generally should not need this argument; instead set the \code{PORTKEY_API_K
 The best place to set this is in \code{.Renviron},
 which you can easily edit by calling \code{usethis::edit_r_environ()}.}
 
-\item{virtual_key}{A virtual identifier storing LLM provider's API key. See
-\href{https://portkey.ai/docs/product/ai-gateway/virtual-keys}{documentation}.
-Can be read from the \code{PORTKEY_VIRTUAL_KEY} environment variable.}
+\item{virtual_key}{\ifelse{html}{\href{https://lifecycle.r-lib.org/articles/stages.html#deprecated}{\figure{lifecycle-deprecated.svg}{options: alt='[Deprecated]'}}}{\strong{[Deprecated]}}.
+Portkey now recommend supplying the model provider
+(formerly known as the \code{virtual_key}), in the model name, e.g.
+\verb{@my-provider/my-model}. See
+\url{https://portkey.ai/docs/support/upgrade-to-model-catalog} for details.
 
-\item{model}{The model to use for the chat (defaults to "gpt-4o").
-We regularly update the default, so we strongly recommend explicitly specifying a model for anything other than casual use.
-Use \code{models_openai()} to see all options.}
+For backward compatibility, the \code{PORTKEY_VIRTUAL_KEY} env var is still used
+if the model
+you can specify the virtual key here.}
 
 \item{params}{Common model parameters, usually created by \code{\link[=params]{params()}}.}
 
@@ -70,16 +70,11 @@ A \link{Chat} object.
 \description{
 \href{https://portkey.ai/docs/product/ai-gateway/universal-api}{PortkeyAI}
 provides an interface (AI Gateway) to connect through its Universal API to a
-variety of LLMs providers with a single endpoint.
-\subsection{Authentication}{
-
-API keys together with configurations of LLM providers are
-stored inside Portkey application.
-}
+variety of LLMs providers via a single endpoint.
 }
 \examples{
 \dontrun{
-chat <- chat_portkey(virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"))
+chat <- chat_portkey()
 chat$chat("Tell me three jokes about statisticians")
 }
 }
diff --git a/plans/chat-openai-compatible.md b/plans/chat-openai-compatible.md
new file mode 100644
index 00000000..bd34c5c2
--- /dev/null
+++ b/plans/chat-openai-compatible.md
@@ -0,0 +1,271 @@
+# Plan: Rename chat_openai to chat_openai_compatible
+
+## Overview
+Rename `chat_openai()` → `chat_openai_compatible()` and `chat_openai_responses()` → `chat_openai()` to better reflect their purposes. The compatible version is for OpenAI-compatible APIs (not official OpenAI), while the main version uses OpenAI's responses API.
+
+## Design Decisions
+1. **Hard rename** - No deprecation period, breaking change
+2. **Remove default base_url** - `chat_openai_compatible()` will require explicit `base_url` parameter
+3. **Keep credentials-only API** - New `chat_openai()` won't add `api_key` parameter
+4. **Rename test helpers** - Update `_test()` functions to match new names
+
+## Implementation Steps
+
+### 1. Core Function Refactoring
+
+#### R/provider-openai.R → R/provider-openai-compatible.R
+- **Rename file**: `provider-openai.R` → `provider-openai-compatible.R`
+- Rename `chat_openai()` → `chat_openai_compatible()`
+- Rename `chat_openai_test()` → `chat_openai_compatible_test()`
+- Rename class: `ProviderOpenAI` → `ProviderOpenAICompatible`
+- Remove default value from `base_url` parameter (make it required)
+- Update documentation:
+  - Title: "Chat with OpenAI-compatible API"
+  - Description: Emphasize this is for OpenAI-compatible APIs (Ollama, vLLM, etc.), NOT for official OpenAI
+  - Add note directing users to `chat_openai()` for official OpenAI API
+  - Update `base_url` parameter docs to indicate it's required
+- Update `@family chatbots` tag references
+
+#### R/provider-openai-responses.R → R/provider-openai.R
+- **Rename file**: `provider-openai-responses.R` → `provider-openai.R`
+- Rename `chat_openai_responses()` → `chat_openai()`
+- Rename `chat_openai_responses_test()` → `chat_openai_test()`
+- Rename class: `ProviderOpenAIResponses` → `ProviderOpenAI`
+- Update documentation:
+  - Title: "Chat with OpenAI"
+  - Description: Main function for official OpenAI API using responses endpoint
+  - Keep all parameters as-is (no `api_key` parameter)
+
+#### R/provider-openai-models.R (if exists)
+- Update `models_openai()` documentation if it references the renamed functions
+
+### 2. Provider Files Using @inheritParams
+
+**19 provider files** use `@inheritParams chat_openai` - update to `@inheritParams chat_openai_compatible`:
+- R/provider-vllm.R - Update wrapper description
+- R/provider-snowflake.R
+- R/provider-portkey.R
+- R/provider-perplexity.R - Update wrapper description
+- R/provider-openrouter.R
+- R/provider-ollama.R - Update wrapper description
+- R/provider-mistral.R
+- R/provider-huggingface.R - Update wrapper description
+- R/provider-groq.R - Update wrapper description
+- R/provider-github.R
+- R/provider-deepseek.R
+- R/provider-databricks.R
+- R/provider-cloudflare.R
+- R/provider-azure.R
+- R/provider-aws.R
+- R/provider-any.R
+
+**Note:** Some providers (vLLM, Perplexity, Ollama, Huggingface, Groq) describe themselves as "lightweight wrapper around chat_openai()" - update to say "Uses OpenAI compatible API via `chat_openai_compatible()`"
+
+### 3. Provider Files Using @inherit return
+
+**17 provider files** use `@inherit chat_openai return` - keep as is since return type is the same.
+
+### 4. Test Files
+
+Update all test files using the renamed functions:
+
+#### tests/testthat/test-provider-openai.R → tests/testthat/test-provider-openai-compatible.R
+- **Rename file**: `test-provider-openai.R` → `test-provider-openai-compatible.R`
+- Rename all `chat_openai_test()` → `chat_openai_compatible_test()` (11 instances)
+- Update test descriptions/context as needed
+
+#### tests/testthat/test-provider-openai-responses.R → tests/testthat/test-provider-openai.R
+- **Rename file**: `test-provider-openai-responses.R` → `test-provider-openai.R`
+- Rename all `chat_openai_responses_test()` → `chat_openai_test()` (10 instances)
+- Update test descriptions/context as needed
+
+#### Other test files using chat_openai_test()
+- tests/testthat/test-parallel-chat.R
+- tests/testthat/test-batch-chat.R
+- tests/testthat/test-content.R
+- tests/testthat/test-chat.R
+- tests/testthat/test-chat-tools.R
+- tests/testthat/test-chat-structured.R
+- tests/testthat/test-utils-auth.R
+- tests/testthat/test-chat-utils.R
+
+**Decision:** These tests should continue to use `chat_openai_test()` so they use the new responses API.
+
+#### Snapshot files
+These will auto-update when tests run:
+- tests/testthat/_snaps/provider-openai.md → tests/testthat/_snaps/provider-openai-compatible.md
+- tests/testthat/_snaps/provider-openai-responses.md → tests/testthat/_snaps/provider-openai.md
+- tests/testthat/_snaps/utils-auth.md
+- tests/testthat/_snaps/chat.md
+- tests/testthat/_snaps/content-replay.md
+- tests/testthat/_snaps/batch-chat.md
+
+### 5. Vignettes
+
+Should not need changes; continue to use `chat_openai()` in order to use latest API.
+
+### 6. README Files
+
+#### README.Rmd (4 usages)
+- Update to `chat_openai()` (responses API)
+- Use `devtools::build_readme()` to update rendered .md
+
+### 7. Source Code Documentation
+
+Update function documentation and examples:
+
+#### R/live.R
+- Update `@param chat` documentation mentioning `chat_openai()`
+
+#### R/content-image.R
+- Update examples using `chat_openai()`
+
+#### R/batch-chat.R
+- Update documentation mentioning `chat_openai()`
+
+#### R/parallel-chat.R
+- Update documentation mentioning `chat_openai()`
+
+#### R/tools-def-auto.R
+- Update `create_tool_def()` defaults using `chat_openai()`
+
+### 8. Man Pages
+
+These will be regenerated by `devtools::document()`
+
+### 9. NAMESPACE
+
+Update exports:
+- Remove: `export(chat_openai)` (OLD meaning)
+- Remove: `export(chat_openai_responses)`
+- Add: `export(chat_openai)` (NEW meaning - responses API)
+- Add: `export(chat_openai_compatible)`
+
+Note: `devtools::document()` should handle this automatically.
+
+### 10. NEWS.md
+
+Add entry at top for next version:
+```markdown
+* `chat_openai()` has been renamed to `chat_openai_compatible()` to clarify it's for OpenAI-compatible APIs, not the official OpenAI API. The `base_url` parameter is now required (#801).
+* `chat_openai_responses()` has been renamed to `chat_openai()`. This is now the main function for using OpenAI's official API via the responses endpoint.
+```
+
+### 11. _pkgdown.yml
+
+No work needed.
+
+### 12. Class Inheritance Considerations
+
+Rename classes to match new APIs:
+- `ProviderOpenAI` → `ProviderOpenAICompatible`
+- `ProviderOpenAIResponses` → `ProviderOpenAI`
+
+This affects all providers that inherit from `ProviderOpenAI`:
+- Update their `parent = ProviderOpenAI` to `parent = ProviderOpenAICompatible`
+- Providers affected: Groq, Ollama, Perplexity, Huggingface, and any others that currently inherit from ProviderOpenAI
+
+## Testing Strategy
+
+1. Run `devtools::document()` to regenerate man pages and NAMESPACE
+2. Run `devtools::test(reporter = "check")` to run all tests
+3. Update snapshot files as needed
+4. Run `pkgdown::check_pkgdown()` to verify all topics in reference
+5. Verify examples in vignettes still work
+6. Check that cross-references work correctly
+
+The vcr cassettes will fail; but Hadley will take care of that.
+
+## Files to Change Summary
+
+### Core Implementation (4 files)
+- R/provider-openai.R → R/provider-openai-compatible.R
+- R/provider-openai-responses.R → R/provider-openai.R
+- tests/testthat/test-provider-openai.R → tests/testthat/test-provider-openai-compatible.R
+- tests/testthat/test-provider-openai-responses.R → tests/testthat/test-provider-openai.R
+
+### Provider Documentation (19 files)
+- All files using `@inheritParams chat_openai` or referencing the function
+
+### Test Files (9+ files)
+- test-provider-openai.R → test-provider-openai-compatible.R
+- test-provider-openai-responses.R → test-provider-openai.R
+- test-parallel-chat.R
+- test-batch-chat.R
+- test-content.R
+- test-chat.R
+- test-chat-tools.R
+- test-chat-structured.R
+- test-utils-auth.R
+- test-chat-utils.R
+
+### Snapshot Files (2 files)
+- _snaps/provider-openai.md → _snaps/provider-openai-compatible.md
+- _snaps/provider-openai-responses.md → _snaps/provider-openai.md
+
+### Documentation Files (6 vignettes + 1 README)
+- vignettes/ellmer.Rmd
+- vignettes/tool-calling.Rmd
+- vignettes/streaming-async.Rmd
+- vignettes/programming.Rmd
+- vignettes/prompt-design.Rmd
+- vignettes/structured-data.Rmd
+- README.Rmd
+
+### Source Code Docs (5 files)
+- R/live.R
+- R/content-image.R
+- R/batch-chat.R
+- R/parallel-chat.R
+- R/tools-def-auto.R
+
+### Other (1 file)
+- NEWS.md
+
+## Estimated Total Files: ~45 files
+
+## Risks and Mitigations
+
+### High Risk
+1. **Breaking change for users** - All code using `chat_openai()` will break
+   - Mitigation: Clear NEWS.md entry, consider blog post/announcement
+
+2. **Documentation inheritance cascade** - 19 files use `@inheritParams`
+   - Mitigation: Careful find-replace, thorough testing of `devtools::document()`
+
+### Medium Risk
+1. **Test disruption** - Many tests use `chat_openai_test()`
+   - Mitigation: Update systematically, run tests frequently
+
+2. **Snapshot mismatches** - Function names in snapshots will change
+   - Mitigation: Review and approve snapshot updates carefully
+
+### Low Risk
+1. **Cross-reference links** - `@family chatbots` should auto-update
+   - Mitigation: Verify after `devtools::document()`
+
+## Open Questions
+
+1. Should general tests use `chat_openai_test()` (responses API) or `chat_openai_compatible_test()`?
+   - **Recommendation:** Use `chat_openai_test()` (responses API) as the primary test function since it's the main OpenAI interface
+
+2. Should we update `OPENAI_BASE_URL` environment variable logic in the compatible version?
+   - Current behavior: Falls back to `OPENAI_BASE_URL` if set
+   - Keep this in `chat_openai_compatible()` but drop from `chat_openai()`
+
+3. How to handle the `service_tier` parameter that's unique to responses API?
+   - It's only in `chat_openai()` (responses), not in `chat_openai_compatible()`
+   - No work needed
+
+## Success Criteria
+
+- [ ] All functions renamed correctly
+- [ ] `base_url` parameter is required in `chat_openai_compatible()`
+- [ ] Documentation clearly distinguishes compatible API vs official OpenAI API
+- [ ] All tests pass
+- [ ] All vignettes knit successfully
+- [ ] `pkgdown::check_pkgdown()` passes
+- [ ] README.md updated
+- [ ] NEWS.md entry added
+- [ ] No broken cross-references in documentation
+- [ ] Snapshot tests updated and reviewed
diff --git a/tests/testthat/_snaps/provider-portkey.md b/tests/testthat/_snaps/provider-portkey.md
deleted file mode 100644
index 0af59bb3..00000000
--- a/tests/testthat/_snaps/provider-portkey.md
+++ /dev/null
@@ -1,7 +0,0 @@
-# defaults are reported
-
-    Code
-      . <- chat_portkey()
-    Message
-      Using model = "gpt-4o".
-
diff --git a/tests/testthat/test-provider-portkey.R b/tests/testthat/test-provider-portkey.R
index 870d4763..498cf34c 100644
--- a/tests/testthat/test-provider-portkey.R
+++ b/tests/testthat/test-provider-portkey.R
@@ -1,53 +1,33 @@
 # Getting started --------------------------------------------------------
 
 test_that("can make simple request", {
-  chat <- chat_portkey_test(
-    virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"),
-    "Be as terse as possible; no punctuation"
-  )
+  chat <- chat_portkey_test("Be as terse as possible; no punctuation")
   resp <- chat$chat("What is 1 + 1?", echo = FALSE)
   expect_match(resp, "2")
   expect_equal(unname(chat$last_turn()@tokens[1:2] > 0), c(TRUE, TRUE))
 })
 
 test_that("can make simple streaming request", {
-  chat <- chat_portkey_test(
-    virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"),
-    "Be as terse as possible; no punctuation"
-  )
+  chat <- chat_portkey_test("Be as terse as possible; no punctuation")
   resp <- coro::collect(chat$stream("What is 1 + 1?"))
   expect_match(paste0(unlist(resp), collapse = ""), "2")
 })
 
 # Common provider interface -----------------------------------------------
 
-test_that("defaults are reported", {
-  expect_snapshot(. <- chat_portkey())
-})
-
 test_that("supports tool calling", {
-  chat_fun <- \(...) {
-    chat_portkey_test(virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"))
-  }
-
+  chat_fun <- chat_portkey_test
   test_tools_simple(chat_fun)
 })
 
 test_that("can extract data", {
-  chat_fun <- \(...) {
-    chat_portkey_test(virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"))
-  }
-
+  chat_fun <- chat_portkey_test
   test_data_extraction(chat_fun)
 })
 
 test_that("can use images", {
   chat_fun <- \(...) {
-    chat_portkey_test(
-      virtual_key = Sys.getenv("PORTKEY_VIRTUAL_KEY"),
-      model = "gpt-4.1-mini",
-      ...
-    )
+    chat_portkey_test(model = "@open-ai-virtual-7f0dcd/gpt-4.1-mini", ...)
   }
 
   test_images_inline(chat_fun)