[Bug]: gpt-oss Intermittent 500 Internal Server Error with empty response body when using strict JSON “function router” system prompt

[bkdoeng]

Your current environment

The output of python collect_env.py

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : version 4.1.0
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.7.1+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.11 (main, Jun  4 2025, 08:56:18) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.2.0-37-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.93
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : GPU 0: NVIDIA H100 80GB HBM3
Nvidia driver version        : 535.129.03
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Address sizes:                      46 bits physical, 48 bits virtual
Byte Order:                         Little Endian
CPU(s):                             16
On-line CPU(s) list:                0-15
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8458P
CPU family:                         6
Model:                              143
Thread(s) per core:                 1
Core(s) per socket:                 16
Socket(s):                          1
Stepping:                           8
BogoMIPS:                           5400.04
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush acpi mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single intel_ppin ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx512_bf16 avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote md_clear flush_l1d arch_capabilities
Hypervisor vendor:                  Xen
Virtualization type:                full
L1d cache:                          768 KiB (16 instances)
L1i cache:                          512 KiB (16 instances)
L2 cache:                           32 MiB (16 instances)
L3 cache:                           1.3 GiB (16 instances)
NUMA node(s):                       1
NUMA node0 CPU(s):                  0-15
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.3.14
[pip3] nvidia-cuda-cupti-cu12==12.8.57
[pip3] nvidia-cuda-nvrtc-cu12==12.8.61
[pip3] nvidia-cuda-runtime-cu12==12.8.57
[pip3] nvidia-cudnn-cu12==9.7.1.26
[pip3] nvidia-cufft-cu12==11.3.3.41
[pip3] nvidia-cufile-cu12==1.13.0.11
[pip3] nvidia-curand-cu12==10.3.9.55
[pip3] nvidia-cusolver-cu12==11.7.2.55
[pip3] nvidia-cusparse-cu12==12.5.7.53
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.8.61
[pip3] nvidia-nvtx-cu12==12.8.55
[pip3] pyzmq==27.0.1
[pip3] torch==2.7.1+cu128
[pip3] torchaudio==2.7.1+cu128
[pip3] torchvision==0.22.1+cu128
[pip3] transformers==4.55.2
[pip3] triton==3.3.1
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.10.1.1
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
  	GPU0	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	0-15	0		N/A

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=all
NVIDIA_REQUIRE_CUDA=cuda>=12.8 brand=unknown,driver>=470,driver<471 brand=grid,driver>=470,driver<471 brand=tesla,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=vapps,driver>=470,driver<471 brand=vpc,driver>=470,driver<471 brand=vcs,driver>=470,driver<471 brand=vws,driver>=470,driver<471 brand=cloudgaming,driver>=470,driver<471 brand=unknown,driver>=535,driver<536 brand=grid,driver>=535,driver<536 brand=tesla,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536 brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536 brand=cloudgaming,driver>=535,driver<536 brand=unknown,driver>=550,driver<551 brand=grid,driver>=550,driver<551 brand=tesla,driver>=550,driver<551 brand=nvidia,driver>=550,driver<551 brand=quadro,driver>=550,driver<551 brand=quadrortx,driver>=550,driver<551 brand=nvidiartx,driver>=550,driver<551 brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551 brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551 brand=cloudgaming,driver>=550,driver<551 brand=unknown,driver>=560,driver<561 brand=grid,driver>=560,driver<561 brand=tesla,driver>=560,driver<561 brand=nvidia,driver>=560,driver<561 brand=quadro,driver>=560,driver<561 brand=quadrortx,driver>=560,driver<561 brand=nvidiartx,driver>=560,driver<561 brand=vapps,driver>=560,driver<561 brand=vpc,driver>=560,driver<561 brand=vcs,driver>=560,driver<561 brand=vws,driver>=560,driver<561 brand=cloudgaming,driver>=560,driver<561 brand=unknown,driver>=565,driver<566 brand=grid,driver>=565,driver<566 brand=tesla,driver>=565,driver<566 brand=nvidia,driver>=565,driver<566 brand=quadro,driver>=565,driver<566 brand=quadrortx,driver>=565,driver<566 brand=nvidiartx,driver>=565,driver<566 brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566 brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566 brand=cloudgaming,driver>=565,driver<566
NCCL_VERSION=2.25.1-1
NVIDIA_DRIVER_CAPABILITIES=compute,utility
NVIDIA_PRODUCT_NAME=CUDA
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=12.8.1
LD_LIBRARY_PATH=/usr/local/cuda/lib64
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

Serving gpt-oss-20b via vLLM’s OpenAI-compatible Chat Completions intermittently returns HTTP 500 with no response body (Content-Length: 0). This occurs for a minimal request using a strict JSON “function router” system prompt (below)

{
  "model": "gpt-oss-20b",
  "max_tokens": 128,
  "messages": [
    {
      "role": "system",
      "content": "Objective: Answer weather-related user queries by using previously called functions when possible, and only call a new function if it is strictly required.\n\nAvailable Data:\nNo previous functions called\n\nInstructions:\n1. **Strictly Use JSON**: Respond only in JSON format with no extra text, explanations, or notes.\n2. **Use Prior Data First**: Treat the results from all previously called functions as **fully available** for answering the query. Do not re-call a function if its data is already provided.\n3. **Only Call the Next Required Function**:\n   - Select the next function only if specific data needed for the query is missing from the available data.\n   - If all necessary data exists, return `no-function-needed`.\n\nResponse Format:\n- Use the following formats based on the query requirements:\n  - **functions_remaining:<count>**: For ongoing function sequences, where `<count>` is the remaining functions needed after the current one.\n    Example: `functions_remaining:1 {\"name\": \"function_name\", \"input\": {\"parameter\": \"value\"}}`\n  - **functions_remaining:0**: Use this when calling the final function required to complete the query.\n  - **no-function-needed**: Use this if the query can be answered with existing data without any additional function calls.\n\nExamples:\n  - If all necessary data is available from previous calls, respond with:\n    `no-function-needed`\n  - If another function is needed to continue, respond with:\n    `functions_remaining:1 {\"name\": \"next_function_needed\", \"input\": {\"key\": \"value\"}}`\n  - When the current call completes the query:\n    `functions_remaining:0 {\"name\": \"final_function\", \"input\": {\"key\": \"value\"}}`\n\nAvailable Functions:\n[\n  {\n    \"name\": \"get_current_weather\",\n    \"description\": \"Fetches the current weather for a given location and datetime.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\", \"description\": \"City, State/Country (e.g., Austin, TX)\"},\n        \"datetime\": {\"type\": \"string\", \"description\": \"ISO 8601 datetime (e.g., 2025-09-01T09:00:00Z)\"}\n      },\n      \"required\": [\"location\", \"datetime\"]\n    }\n  },\n  {\n    \"name\": \"get_daily_forecast\",\n    \"description\": \"Retrieves a daily forecast for a given location and date.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\", \"description\": \"City, State/Country\"},\n        \"date\": {\"type\": \"string\", \"description\": \"ISO 8601 date (e.g., 2025-09-01)\"}\n      },\n      \"required\": [\"location\", \"date\"]\n    }\n  },\n  {\n    \"name\": \"get_air_quality\",\n    \"description\": \"Fetches air quality for a given location and datetime.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\"},\n        \"datetime\": {\"type\": \"string\", \"description\": \"ISO 8601 datetime\"}\n      },\n      \"required\": [\"location\", \"datetime\"]\n    }\n  }\n]\n\nNOTE: Respond strictly with the required `functions_remaining:<count> <json>` format on a single line, without any additional text."
    },
    {
      "role": "user",
      "content": "What's the weather in Austin, TX on 2025-09-01?"
    }
  ],
  "tools": [],
  "tool_choice": "none"
}

curl -i http://localhost:8000/v1/chat/completions \ -H "Content-Type: application/json" \ --data-binary @payload.json

out of 200 requests 14 got empty response (500 internal server error), in vllm server the is as below

"POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

If possible, I’d love:
• A change so that malformed model output does not produce a 500/empty body (return 200 with raw text or 4xx/5xx with a JSON error object).

To reproduce this issue
Attached python script to run it concurrently 20 with total of 200 requests, below is the command

python gptoss_weather_stress_logged.py -n 200 -c 20 --model "openai/gpt-oss-20b"

Requests with temperature to 0 and top_p as 1 have way less errors and sometimes no errors(rerun the script in that case)

gptoss_weather_stress_logged.py

Thanks in advance!

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[liushz]

Same question

[fabienric]

Same issue here.

Instead of returning a 200 in this specific case of an empty output, I think it would be more interesting to dig why VLLM fails at generating the response in the first place? Is there an underlying error which is not propagated?

The fact that the temperature influences the ratio of 500 errors you get seems to indicate a relationship between generated tokens and the occurrence of this error. Maybe a parsing issue?

[damodaryekkuluri]

same issue.

[dafu-wu]

same issue.

[wonjerry]

same issue. is it related to harmony parser?

[daviden1013]

Same issue. About half of the requests got "500 Internal Server Error".

[etrangerlx]

same issue.

[etrangerlx]

Your current environment

The output of python collect_env.py

🐛 Describe the bug

Serving gpt-oss-20b via vLLM’s OpenAI-compatible Chat Completions intermittently returns HTTP 500 with no response body (Content-Length: 0). This occurs for a minimal request using a strict JSON “function router” system prompt (below)

{
  "model": "gpt-oss-20b",
  "max_tokens": 128,
  "messages": [
    {
      "role": "system",
      "content": "Objective: Answer weather-related user queries by using previously called functions when possible, and only call a new function if it is strictly required.\n\nAvailable Data:\nNo previous functions called\n\nInstructions:\n1. **Strictly Use JSON**: Respond only in JSON format with no extra text, explanations, or notes.\n2. **Use Prior Data First**: Treat the results from all previously called functions as **fully available** for answering the query. Do not re-call a function if its data is already provided.\n3. **Only Call the Next Required Function**:\n   - Select the next function only if specific data needed for the query is missing from the available data.\n   - If all necessary data exists, return `no-function-needed`.\n\nResponse Format:\n- Use the following formats based on the query requirements:\n  - **functions_remaining:<count>**: For ongoing function sequences, where `<count>` is the remaining functions needed after the current one.\n    Example: `functions_remaining:1 {\"name\": \"function_name\", \"input\": {\"parameter\": \"value\"}}`\n  - **functions_remaining:0**: Use this when calling the final function required to complete the query.\n  - **no-function-needed**: Use this if the query can be answered with existing data without any additional function calls.\n\nExamples:\n  - If all necessary data is available from previous calls, respond with:\n    `no-function-needed`\n  - If another function is needed to continue, respond with:\n    `functions_remaining:1 {\"name\": \"next_function_needed\", \"input\": {\"key\": \"value\"}}`\n  - When the current call completes the query:\n    `functions_remaining:0 {\"name\": \"final_function\", \"input\": {\"key\": \"value\"}}`\n\nAvailable Functions:\n[\n  {\n    \"name\": \"get_current_weather\",\n    \"description\": \"Fetches the current weather for a given location and datetime.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\", \"description\": \"City, State/Country (e.g., Austin, TX)\"},\n        \"datetime\": {\"type\": \"string\", \"description\": \"ISO 8601 datetime (e.g., 2025-09-01T09:00:00Z)\"}\n      },\n      \"required\": [\"location\", \"datetime\"]\n    }\n  },\n  {\n    \"name\": \"get_daily_forecast\",\n    \"description\": \"Retrieves a daily forecast for a given location and date.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\", \"description\": \"City, State/Country\"},\n        \"date\": {\"type\": \"string\", \"description\": \"ISO 8601 date (e.g., 2025-09-01)\"}\n      },\n      \"required\": [\"location\", \"date\"]\n    }\n  },\n  {\n    \"name\": \"get_air_quality\",\n    \"description\": \"Fetches air quality for a given location and datetime.\",\n    \"parameters\": {\n      \"type\": \"object\",\n      \"properties\": {\n        \"location\": {\"type\": \"string\"},\n        \"datetime\": {\"type\": \"string\", \"description\": \"ISO 8601 datetime\"}\n      },\n      \"required\": [\"location\", \"datetime\"]\n    }\n  }\n]\n\nNOTE: Respond strictly with the required `functions_remaining:<count> <json>` format on a single line, without any additional text."
    },
    {
      "role": "user",
      "content": "What's the weather in Austin, TX on 2025-09-01?"
    }
  ],
  "tools": [],
  "tool_choice": "none"
}

curl -i http://localhost:8000/v1/chat/completions \ -H "Content-Type: application/json" \ --data-binary @payload.json

out of 200 requests 14 got empty response (500 internal server error), in vllm server the is as below

"POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

If possible, I’d love: • A change so that malformed model output does not produce a 500/empty body (return 200 with raw text or 4xx/5xx with a JSON error object).

To reproduce this issue Attached python script to run it concurrently 20 with total of 200 requests, below is the command

python gptoss_weather_stress_logged.py -n 200 -c 20 --model "openai/gpt-oss-20b"

Requests with temperature to 0 and top_p as 1 have way less errors and sometimes no errors(rerun the script in that case)

gptoss_weather_stress_logged.py

Thanks in advance!

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

max_tokens should be set 1024,4096?

[FrankTheTank9]

We are having the same issue with gpt-oss-120b. Since the output is empty, this is hard to know exact cause of the error and application code cannot try to deal with the error.

[tamastarjanyi]

We also have the same issue both with gpt-oss-120b and gpt-oss-20b. This is somehow prompt specific. Many prompts work fine but some prompts are causing http 500 issue. But not always. The same "bad" prompt sometimes works sometimes does not.
Moreover, if you set reasoning_effort to low error rate is the highest (5-6% to me). If you set reasoning_effort to medium it is better (2-3% error rate). And in my case if I set reasoning_effort to high I get no error anymore. (0% error) If somebody can try it and confirm this, probably this would help identifying the issue. But so far this is my experience.

[FrankTheTank9]

We also have the same issue both with gpt-oss-120b and gpt-oss-20b. This is somehow prompt specific. Many prompts work fine but some prompts are causing http 500 issue. But not always. The same "bad" prompt sometimes works sometimes does not. Moreover, if you set reasoning_effort to low error rate is the highest (5-6% to me). If you set reasoning_effort to medium it is better (2-3% error rate). And in my case if I set reasoning_effort to high I get no error anymore. (0% error) If somebody can try it and confirm this, probably this would help identifying the issue. But so far this is my experience.

We see similar behavior. Is your understanding that the 500 errors happen when the model does not provide valid json? This is our hypothesis but hard to confirm given model does not return response.

[tamastarjanyi]

We also have the same issue both with gpt-oss-120b and gpt-oss-20b. This is somehow prompt specific. Many prompts work fine but some prompts are causing http 500 issue. But not always. The same "bad" prompt sometimes works sometimes does not. Moreover, if you set reasoning_effort to low error rate is the highest (5-6% to me). If you set reasoning_effort to medium it is better (2-3% error rate). And in my case if I set reasoning_effort to high I get no error anymore. (0% error) If somebody can try it and confirm this, probably this would help identifying the issue. But so far this is my experience.

We see similar behavior. Is your understanding that the 500 errors happen when the model does not provide valid json? This is our hypothesis but hard to confirm given model does not return response.

I don't think so. I have two prompts generating this issue at the moment. (Unfortunately, they are proprietary, and I cannot share those but only one of those requires json output. The other is detecting user intent. And actually, both are used to detect user intent somehow. [Intent examples are given in the system prompt.])
So, I cannot say the issue is not json related but in one of my prompts I do not ask the model to generate json output in any way. Neither in the text nor in the system prompt not even in any parameter. Still the model might belive so.
And since I have seen issue is not happening when the reasoning_effort is set to high (default is medium in the default chat template) this might be something related to response parsing? But this is just an idea at the moment.

[FrankTheTank9]

We also have the same issue both with gpt-oss-120b and gpt-oss-20b. This is somehow prompt specific. Many prompts work fine but some prompts are causing http 500 issue. But not always. The same "bad" prompt sometimes works sometimes does not. Moreover, if you set reasoning_effort to low error rate is the highest (5-6% to me). If you set reasoning_effort to medium it is better (2-3% error rate). And in my case if I set reasoning_effort to high I get no error anymore. (0% error) If somebody can try it and confirm this, probably this would help identifying the issue. But so far this is my experience.

We see similar behavior. Is your understanding that the 500 errors happen when the model does not provide valid json? This is our hypothesis but hard to confirm given model does not return response.

I don't think so. I have two prompts generating this issue at the moment. (Unfortunately, they are proprietary, and I cannot share those but only one of those requires json output. The other is detecting user intent. And actually, both are used to detect user intent somehow. [Intent examples are given in the system prompt.]) So, I cannot say the issue is not json related but in one of my prompts I do not ask the model to generate json output in any way. Neither in the text nor in the system prompt not even in any parameter. Still the model might belive so. And since I have seen issue is not happening when the reasoning_effort is set to high (default is medium in the default chat template) this might be something related to response parsing? But this is just an idea at the moment.

Thanks. We are looking to try with response api instead of chat api to see if this improves as currently a blocker for us. Is this something you have tried?

[tamastarjanyi]

We also have the same issue both with gpt-oss-120b and gpt-oss-20b. This is somehow prompt specific. Many prompts work fine but some prompts are causing http 500 issue. But not always. The same "bad" prompt sometimes works sometimes does not. Moreover, if you set reasoning_effort to low error rate is the highest (5-6% to me). If you set reasoning_effort to medium it is better (2-3% error rate). And in my case if I set reasoning_effort to high I get no error anymore. (0% error) If somebody can try it and confirm this, probably this would help identifying the issue. But so far this is my experience.

We see similar behavior. Is your understanding that the 500 errors happen when the model does not provide valid json? This is our hypothesis but hard to confirm given model does not return response.

I don't think so. I have two prompts generating this issue at the moment. (Unfortunately, they are proprietary, and I cannot share those but only one of those requires json output. The other is detecting user intent. And actually, both are used to detect user intent somehow. [Intent examples are given in the system prompt.]) So, I cannot say the issue is not json related but in one of my prompts I do not ask the model to generate json output in any way. Neither in the text nor in the system prompt not even in any parameter. Still the model might belive so. And since I have seen issue is not happening when the reasoning_effort is set to high (default is medium in the default chat template) this might be something related to response parsing? But this is just an idea at the moment.

Thanks. We are looking to try with response api instead of chat api to see if this improves as currently a blocker for us. Is this something you have tried?

I actually was doing that meanwhile. And on my side the result is a json even when I am not requesting that AND with /v1/responses API there is NO ERROR.

So it seems the issue is in the response parsing on the /v1/chat/completions api. Let's see if you have the same result.