From f0e3718222ba331ce5d027d5a75bb5375e1db9ad Mon Sep 17 00:00:00 2001
From: Nikitha Suryadevara <nsuryadevara@together.ai>
Date: Wed, 21 Jan 2026 12:07:45 -0800
Subject: [PATCH] feat(cli): add b200 and h200 GPU options for endpoint
 creation

---
 src/together/cli/api/endpoints.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/together/cli/api/endpoints.py b/src/together/cli/api/endpoints.py
index bb2a9eb..97272b7 100644
--- a/src/together/cli/api/endpoints.py
+++ b/src/together/cli/api/endpoints.py
@@ -98,7 +98,7 @@ def endpoints(ctx: click.Context) -> None:
 )
 @click.option(
     "--gpu",
-    type=click.Choice(["h100", "a100", "l40", "l40s", "rtx-6000"]),
+    type=click.Choice(["b200", "h200", "h100", "a100", "l40", "l40s", "rtx-6000"]),
     required=True,
     help="GPU type to use for inference",
 )
@@ -161,6 +161,8 @@ def create(
     """Create a new dedicated inference endpoint."""
     # Map GPU types to their full hardware ID names
     gpu_map = {
+        "b200": "nvidia_b200_180gb_sxm",
+        "h200": "nvidia_h200_140gb_sxm",
         "h100": "nvidia_h100_80gb_sxm",
         "a100": "nvidia_a100_80gb_pcie" if gpu_count == 1 else "nvidia_a100_80gb_sxm",
         "l40": "nvidia_l40",