Critical browser memory/CPU leak when returning a new component from `Blocks.load()` with 3+ `Tab`s under a `Column` (regression in 6.11.0+)

[tripleS-Dev]

Describe the bug

When a Gradio app is structured with a Column containing 3 or more Tabs, and a component such as Textbox, Radio, or Dropdown exists under that same Column, returning a new component instance from Blocks.load() causes a critical browser-side failure.

In my environment, this triggers an infinite loop of frontend errors in the browser console, RAM usage rises to 100%, and eventually CPU usage also reaches 100%. This makes the page effectively unusable for end users.

This appears to be a regression:

• Gradio 6.3.0: no issue
• Gradio 6.10.0: no issue
• Gradio 6.11.0: issue occurs
• Gradio 6.12.0: issue occurs

So the bug appears to have been introduced in 6.11.0 or later. Based on my testing, versions <= 6.10.0 do not seem to have this problem.

The reproduction may look unusual, but every line and condition in the example is required to reproduce the issue consistently.

*I wrote all of the content in this document myself, and only used chatgpt to improve readability (Markdown).

Have you searched existing issues? 🔎

• I have searched and found no existing issues

Reproduction

import gradio as gr

with gr.Blocks() as dem:
    with gr.Column():

        with gr.Tab("tab1"):
            with gr.Column():
                test_comp = gr.Textbox()

        with gr.Tab("tab2"):
            pass

        with gr.Tab("tab3"):
            pass

    dem.load(fn=lambda: gr.Textbox(), outputs=test_comp)

dem.launch()

This issue reproduces only when all of the following conditions are met simultaneously:

1. The Tabs must be placed under a Column.

2. A component such as gr.Radio(), gr.Dropdown(), or gr.Textbox() must also exist under that Column.
   These are only examples; the same issue also occurs with other components.

3. The issue only occurs when there are 3 or more Tabs under the Column.

4. The .load() callback must return a component instance, for example:

   dem.load(fn=lambda: gr.Checkbox(), outputs=test_comp)

If any of these conditions is removed, the issue does not reproduce in my testing.

Screenshot

No response

Logs

The following error is repeated infinitely in the browser console:


Cmnr_FP2.js:14 Uncaught
Di @ index-Cmnr_FP2.js:14
uo @ index-Cmnr_FP2.js:45
Pr @ index-Cmnr_FP2.js:45
flush @ index-Cmnr_FP2.js:45
(anonymous) @ index-Cmnr_FP2.js:45
Ri @ index-Cmnr_FP2.js:4
$r @ index-Cmnr_FP2.js:39
(anonymous) @ index-Cmnr_FP2.js:39

index-Cmnr_FP2.js:14 Uncaught
Di @ index-Cmnr_FP2.js:14
uo @ index-Cmnr_FP2.js:45
Pr @ index-Cmnr_FP2.js:45
flush @ index-Cmnr_FP2.js:45
(anonymous) @ index-Cmnr_FP2.js:45
Ri @ index-Cmnr_FP2.js:4
$r @ index-Cmnr_FP2.js:39
(anonymous) @ index-Cmnr_FP2.js:39


This repeats endlessly and causes severe browser resource exhaustion.

System Info

* OS: Windows
* Browser: Chrome and other Chromium-based browsers
* Gradio version: 

Gradio Environment Information:
------------------------------
Operating System: Windows
gradio version: 6.12.0
gradio_client version: 2.4.1

------------------------------------------------
gradio dependencies in your environment:

anyio: 4.9.0
audioop-lts is not installed.
brotli: 1.1.0
fastapi: 0.115.12
gradio-client: 2.4.1
groovy: 0.1.2
hf-gradio: 0.4.1
httpx: 0.28.1
huggingface-hub: 1.2.3
jinja2: 3.1.6
markupsafe: 3.0.2
numpy: 2.2.6
orjson: 3.10.18
packaging: 25.0
pandas: 2.3.0
pillow: 11.2.1
pydantic: 2.11.10
pydub: 0.25.1
python-multipart: 0.0.20
pytz: 2026.1.post1
pyyaml: 6.0.2
safehttpx: 0.1.7
semantic-version: 2.10.0
starlette: 0.46.2
tomlkit: 0.13.3
typer: 0.16.0
typing-extensions: 4.15.0
uvicorn: 0.34.3
mcp is not installed.
pydantic: 2.11.10
authlib is not installed.
itsdangerous is not installed.


gradio_client dependencies in your environment:

fsspec: 2025.5.1
httpx: 0.28.1
huggingface-hub: 1.2.3
packaging: 25.0
typing-extensions: 4.15.0

Severity

Blocking usage of gradio

[NoerNova]

Same issue occurs on gradio-6.13.0
Downgrading to version 6.10.0 fixed the issue.

[tripleS-Dev]

I haven’t minimized it, but here’s another example from my project where the same issue occurred. When I click
go_next.click(fn=lambda : gr.Walkthrough(selected=1), outputs=walkthrough)
, the same memory leak and freeze occur.

import gradio as gr


with gr.Blocks() as demo:
    with gr.Walkthrough() as walkthrough:
        with gr.Step("Upload/Extract", id=0):
            with gr.Column():
                input_pdf = gr.File(type='filepath', file_count='single', file_types=['.pdf'], interactive=True)
                input_select = gr.Textbox(interactive=True)

                with gr.Row():
                    extract_btn = gr.Button(value='Extract', variant='secondary', interactive=True)
                    go_next = gr.Button(value='Next', variant='secondary', interactive=True)
                    go_next.click(fn=lambda : gr.Walkthrough(selected=1), outputs=walkthrough)

        with gr.Step("Template", id=1):
            with gr.Tabs():
                with gr.Tab('PROP AXIS'):
                    with gr.Group():
                        with gr.Row():
                            main_title = gr.Textbox(value='2027학년도 S1_H1', interactive=True, label='Main Title', min_width=1)
                            template_name = gr.Textbox(value='PROP AXIS', interactive=True, label='Template Name', min_width=1)
                        with gr.Row():
                            sub_title = gr.Textbox(value='- SS-09 -', interactive=True, label='Sub Title', min_width=1)
                            question_start_name = gr.Textbox(value='선택형 문항', interactive=True, label='문항 종류', min_width=1)

                    generate_btn = gr.Button(variant='secondary', interactive=True)

                    preview = gr.Gallery(interactive=False, label='Preview')

                    output_file = gr.File(interactive=False)

                with gr.Tab('CSAT'):
                    pass



if __name__ == "__main__":
    print('http://localhost:7860')
    demo.launch()