From 3ea4b715ba8d6c1666205443a12475d74a1af040 Mon Sep 17 00:00:00 2001
From: siyoon <siyoon@friendli.ai>
Date: Sun, 30 Aug 2026 14:22:41 +0900
Subject: [PATCH] feat(friendli): add zai-org/GLM-5.3-Flash model pricing

Per https://api.friendli.ai/serverless/v1/models:
- $0.15 input / $0.50 output / $0.03 cached input per MTok
- 1M context, 1M max output, reasoning with effort low/high/max
  (per HF chat_template.jinja: low/high honored, anything else -> max)
- tool calling, parallel tool calls, structured output, prompt caching
- image + video input (native multimodal)
---
 model_prices_and_context_window.json          | 25 +++++++++++++
 ...t_friendli_glm_5_3_flash_model_metadata.py | 37 +++++++++++++++++++
 2 files changed, 62 insertions(+)
 create mode 100644 tests/test_litellm/test_friendli_glm_5_3_flash_model_metadata.py

diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index 05c1cfd3179..fd6b9e11adf 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -19564,6 +19564,31 @@
         "supports_system_messages": true,
         "supports_tool_choice": true
     },
+    "friendliai/zai-org/GLM-5.3-Flash": {
+        "litellm_provider": "friendliai",
+        "supports_reasoning": true,
+        "supports_function_calling": true,
+        "max_input_tokens": 1048576,
+        "max_tokens": 1048576,
+        "max_output_tokens": 1048576,
+        "input_cost_per_token": 1.5e-07,
+        "output_cost_per_token": 5e-07,
+        "cache_read_input_token_cost": 3e-08,
+        "supports_prompt_caching": true,
+        "supports_max_reasoning_effort": true,
+        "supports_low_reasoning_effort": true,
+        "supports_parallel_function_calling": true,
+        "supports_response_schema": true,
+        "supports_native_structured_output": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "mode": "chat",
+        "comment": "Native multimodal GLM model for efficient coding and long-horizon agent tasks",
+        "source": "https://api.friendli.ai/serverless/v1/models",
+        "supports_vision": true,
+        "supports_image_input": true,
+        "supports_video_input": true
+    },
     "ft:babbage-002": {
         "deprecation_date": "2026-10-23",
         "input_cost_per_token": 1.6e-06,
diff --git a/tests/test_litellm/test_friendli_glm_5_3_flash_model_metadata.py b/tests/test_litellm/test_friendli_glm_5_3_flash_model_metadata.py
new file mode 100644
index 00000000000..5110307e5bb
--- /dev/null
+++ b/tests/test_litellm/test_friendli_glm_5_3_flash_model_metadata.py
@@ -0,0 +1,37 @@
+import json
+from pathlib import Path
+
+from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
+
+
+def test_friendli_glm_5_3_flash_model_info():
+    model = "friendliai/zai-org/GLM-5.3-Flash"
+    json_path = Path(__file__).parents[2] / "model_prices_and_context_window.json"
+    with open(json_path) as f:
+        model_cost = json.load(f)
+
+    info = model_cost.get(model)
+    assert (
+        info is not None
+    ), f"{model} not found in model_prices_and_context_window.json"
+    assert info["litellm_provider"] == "friendliai"
+    assert info["mode"] == "chat"
+    assert info["input_cost_per_token"] == 1.5e-07
+    assert info["output_cost_per_token"] == 5e-07
+    assert info["cache_read_input_token_cost"] == 3e-08
+    assert info["max_input_tokens"] == 1048576
+    assert info["max_output_tokens"] == 1048576
+    assert info["supports_function_calling"] is True
+    assert info["supports_reasoning"] is True
+    assert info["supports_low_reasoning_effort"] is True
+    assert info["supports_max_reasoning_effort"] is True
+    assert info["supports_tool_choice"] is True
+    assert info["supports_prompt_caching"] is True
+    # Friendli serves image and video input
+    assert info["supports_vision"] is True
+    assert info["supports_image_input"] is True
+    assert info["supports_video_input"] is True
+
+    routed_model, provider, _, _ = get_llm_provider(model=model)
+    assert routed_model == "zai-org/GLM-5.3-Flash"
+    assert provider == "friendliai"