diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index 543864d4a..3bc9b81ad 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -3260,7 +3260,7 @@ dsr1-fp8-h100-dynamo-sglang:
         dp-attn: true
 
 gptoss-fp4-h200-trt:
-  image: nvcr.io#nvidia/tensorrt-llm/release:gpt-oss-dev
+  image: nvcr.io#nvidia/tensorrt-llm/release:1.3.0rc5
   model: openai/gpt-oss-120b
   model-prefix: gptoss
   runner: h200
diff --git a/benchmarks/single_node/gptoss_fp4_h200_trt.sh b/benchmarks/single_node/gptoss_fp4_h200_trt.sh
index f230cd9b0..d4891130c 100644
--- a/benchmarks/single_node/gptoss_fp4_h200_trt.sh
+++ b/benchmarks/single_node/gptoss_fp4_h200_trt.sh
@@ -13,9 +13,6 @@ check_env_vars \
     DP_ATTENTION \
     EP_SIZE
 
-# TensorRT bug. Remove when fixed
-sed -i '417d' /usr/local/lib/python3.12/dist-packages/tensorrt_llm/executor/result.py
-
 if [[ -n "$SLURM_JOB_ID" ]]; then
   echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
 fi
@@ -36,7 +33,6 @@ cuda_graph_config:
 enable_attention_dp: $DP_ATTENTION
 kv_cache_config:
   dtype: auto
-  enable_block_reuse: false
   free_gpu_memory_fraction: 0.85
 moe_config:
   backend: TRITON
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 006522fc8..6b0beb137 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -824,3 +824,12 @@
     - "Uses triton attention backend, TP=8, concurrency 4-64"
     - "Following AMD Andy Luo's recipe"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/850
+  
+- config-keys:
+    - gptoss-fp4-h200-trt
+  description:
+    - "Upgrade TensorRT-LLM container from release:gpt-oss-dev to release:v1.3.0rc5"
+    - "Remove sed hack for TensorRT bug (fixed upstream in v1.3.0rc5)"
+    - "Remove enable_block_reuse: false from kv_cache_config (default true is now recommended)"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/854
+