diff --git a/hugo/content/en/gpu_monitoring/tracing.md b/hugo/content/en/gpu_monitoring/tracing/_index.md
similarity index 90%
rename from hugo/content/en/gpu_monitoring/tracing.md
rename to hugo/content/en/gpu_monitoring/tracing/_index.md
index b287e0a8350..18411dcf193 100644
--- a/hugo/content/en/gpu_monitoring/tracing.md
+++ b/hugo/content/en/gpu_monitoring/tracing/_index.md
@@ -4,6 +4,9 @@ is_beta: true
 private: true
 description: Enable GPU activity tracing for selected Kubernetes workloads.
 further_reading:
+- link: "/gpu_monitoring/tracing/compatibility"
+  tag: "Documentation"
+  text: "GPU tracing compatibility"
 - link: "/gpu_monitoring/setup"
   tag: "Documentation"
   text: "Set up GPU Monitoring"
@@ -12,9 +15,9 @@ further_reading:
   text: "Learn more about what GPU Monitoring offers"
 ---
 
-{{< beta-callout url="#" btn_hidden="true" >}}
+{{< callout url="#" btn_hidden="true" >}}
 Continuous tracing with GPU Monitoring is in Early Access Preview.
-{{< /beta-callout >}}
+{{< /callout >}}
 
 ## Overview
 
@@ -28,7 +31,7 @@ Continuous tracing with GPU Monitoring enables lightweight GPU activity tracing
 
 To begin continuously tracing your workloads, you must first meet the following criteria:
 - You are running the Datadog Cluster Agent version 7.80 or later with [GPU Monitoring enabled][1].
-- Minimum required CUDA and CUPTI version: 13.
+- Minimum required CUDA and CUPTI version: 13. For supported tracer and framework versions, see [GPU tracing compatibility][3].
 
 ### 1. Configure GPU tracing
 
@@ -99,3 +102,4 @@ pod_name:<NEW_GPU_POD> kube_namespace:<GPU_WORKLOAD_NAMESPACE>
 
 [1]: /gpu_monitoring/setup
 [2]: /tracing/trace_explorer/
+[3]: /gpu_monitoring/tracing/compatibility/
diff --git a/hugo/content/en/gpu_monitoring/tracing/compatibility.md b/hugo/content/en/gpu_monitoring/tracing/compatibility.md
new file mode 100644
index 00000000000..90740761a69
--- /dev/null
+++ b/hugo/content/en/gpu_monitoring/tracing/compatibility.md
@@ -0,0 +1,43 @@
+---
+title: GPU Tracing Compatibility
+is_beta: true
+private: true
+description: Review supported CUDA, CUPTI, tracer, and framework versions for GPU tracing.
+further_reading:
+- link: "/gpu_monitoring/tracing"
+  tag: "Documentation"
+  text: "Set up continuous tracing with GPU Monitoring"
+- link: "/gpu_monitoring/setup"
+  tag: "Documentation"
+  text: "Set up GPU Monitoring"
+---
+
+{{< callout url="#" btn_hidden="true" >}}
+<a href="/gpu_monitoring/tracing/">Continuous tracing</a> with GPU Monitoring is in Early Access Preview.
+{{< /callout >}}
+
+GPU tracing supports CUDA 13.x with CUPTI 13.x. CUDA 12.x and earlier are not supported.
+
+## Tracer versions
+
+| C tracer version | CUDA | CUPTI | Framework coverage | Important changes |
+|---|---|---|---|---|
+| 0.18.0 | 13.x | 13.x | PyTorch, vLLM, and TensorRT-LLM | Adds vLLM request correlation and improves profiling efficiency for short workloads. |
+| 0.19.0 | 13.x | 13.x | PyTorch, vLLM, and TensorRT-LLM | Improves startup reliability across PyTorch and CUDA environments. |
+| 0.19.1 | 13.x | 13.x | PyTorch, vLLM, and TensorRT-LLM | Improves completeness and accuracy of NCCL activity metrics. |
+| 0.20.0 | 13.x | 13.x | PyTorch, vLLM, TensorRT-LLM, and SGLang | Adds SGLang inference-step GPU tracing. |
+| 0.21.0 | 13.x | 13.x | PyTorch, vLLM, TensorRT-LLM, and SGLang | Adds CUDA Graph identity to kernel spans and reduces sampling overhead. |
+| 0.21.1 | 13.x | 13.x | PyTorch, vLLM, TensorRT-LLM, and SGLang | Reduces workload overhead and improves GPU tracing reliability across x86_64 and ARM64. **Recommended.** |
+
+## Minimum framework versions
+
+| Framework | Minimum version | C tracer version | Coverage |
+|---|---|---|---|
+| PyTorch | 2.8.0 | 0.18.0+ | Training, inference, CUDA, and NCCL activity |
+| vLLM | 0.11.0 | 0.18.0+ | Prefill and decode boundaries with request correlation |
+| TensorRT-LLM | 1.2.0 | 0.18.0+ | PyTorch executor prefill and decode tracing |
+| SGLang | 0.5.10.post1 | 0.20.0+ | Prefill, decode, and mixed inference tracing |
+
+## Further reading
+
+{{< partial name="whats-next/whats-next.html" >}}