feat(xai): allow reasoning_effort on Grok-4 family

guides/xai.md

@@ -88,6 +88,7 @@ Passed via `:provider_options` keyword:
 - Do NOT support `stop`, `presence_penalty`, or `frequency_penalty`
 - Use `max_completion_tokens` instead of `max_tokens`
 - Support native structured outputs
+- Support the `reasoning_effort` parameter (`"none"`, `"low"`, `"medium"`, `"high"`) — see [xAI's reasoning docs](https://docs.x.ai/developers/model-capabilities/text/reasoning)
 
 ### Grok-3-mini Models
 - Support `reasoning_effort` parameter (`"low"`, `"medium"`, `"high"`)

lib/req_llm/providers/xai.ex

@@ -37,7 +37,7 @@ defmodule ReqLLM.Providers.XAI do
 
   Beyond standard OpenAI parameters, xAI supports:
   - `max_completion_tokens` - Preferred over max_tokens for Grok-4 models
-  - `reasoning_effort` - Reasoning level (low, medium, high) for Grok-3 mini models only
+  - `reasoning_effort` - Reasoning level (low, medium, high, none) for Grok-3 mini and Grok-4 family models
   - `xai_tools` - Agent tools configuration (e.g., web_search, x_search)
   - `parallel_tool_calls` - Allow parallel function calls (default: true)
   - `stream_options` - Streaming configuration (include_usage)
@@ -46,7 +46,7 @@ defmodule ReqLLM.Providers.XAI do
   ## Model Compatibility Notes
 
   - Native structured outputs supported on models >= `grok-2-1212` and `grok-2-vision-1212`
-  - `reasoning_effort` is only supported for grok-3-mini and grok-3-mini-fast models
+  - `reasoning_effort` is supported for grok-3-mini, grok-3-mini-fast, and Grok-4 family models (see https://docs.x.ai/developers/model-capabilities/text/reasoning)
   - Grok-4 models do not support `stop`, `presence_penalty`, or `frequency_penalty`
   - Agent tools (e.g., web_search) incur additional costs per source
 
@@ -713,7 +713,7 @@ defmodule ReqLLM.Providers.XAI do
     {opts, Enum.reverse(warnings)}
   end
 
-  def translate_options(_operation, model, opts) do
+  def translate_options(_operation, _model, opts) do
     warnings = []
 
     {stream_value, opts} = Keyword.pop(opts, :stream?)
@@ -775,18 +775,11 @@ defmodule ReqLLM.Providers.XAI do
 
     {reasoning_effort, opts} = Keyword.pop(opts, :reasoning_effort)
 
-    {opts, warnings} =
+    opts =
       if reasoning_effort do
-        model_name = model.id
-
-        if String.contains?(model_name, "grok-4") do
-          warning = "reasoning_effort is not supported for Grok-4 models and will be ignored"
-          {opts, [warning | warnings]}
-        else
-          {Keyword.put(opts, :reasoning_effort, reasoning_effort), warnings}
-        end
+        Keyword.put(opts, :reasoning_effort, reasoning_effort)
       else
-        {opts, warnings}
+        opts
       end
 
     {opts, Enum.reverse(warnings)}

test/providers/xai_test.exs

@@ -630,12 +630,14 @@ defmodule ReqLLM.Providers.XAITest do
       assert Keyword.get(translated_opts, :reasoning_effort) == "high"
       assert warnings == []
 
+      # Grok-4 family now supports reasoning_effort
+      # (https://docs.x.ai/developers/model-capabilities/text/reasoning) — the
+      # parameter is forwarded to the API rather than dropped with a warning.
       {:ok, grok_4} = ReqLLM.model("xai:grok-4")
       {translated_opts, warnings} = XAI.translate_options(:chat, grok_4, opts)
 
-      refute Keyword.has_key?(translated_opts, :reasoning_effort)
-      assert length(warnings) == 1
-      assert hd(warnings) =~ "Grok-4"
+      assert Keyword.get(translated_opts, :reasoning_effort) == "high"
+      assert warnings == []
     end
   end