[2] progress

match0 / —

unseen—

[3] judge by

taskrequested elements present, convincingly combined

compclear, intentional, readable at a glance

finishmaterials, lighting, detail resolved

[4] filters: all

harnessmodeleffort

identities sealed until vote

[5] task

Loading comparison…

Loading task instructions…

best follows the task · strongest overall quality

[A] candidate

[B] candidate

VOTEmatch —

blind · sides randomized

[2] rank by

advanced ranking modes Model × Effort

[3] scope

[4] / search · filters: all

modelharnesseffortmin sample

[5] rankings

rankings · quality

Quality ranks models by human preference, grouped by reasoning effort.

[7] leaderboard · by quality (elo)

Configuration rankings
			cost/run · tokens	confidence
Loading standings…

Ratings begin at 1500. Model × Effort pools across harnesses and profiles; Harness uses direct same-model, same-effort comparisons. Configurations with fewer than five votes are marked provisional.

[2] totals — total spent · all runs

runs—

matching—

[3] filters

modelharnesseffortprompt

[4] run explorer

runs

search and filter imported runs, then open one to inspect its render, 3d model and recorded metrics.

[5] runs · newest first

Imported run history
run · date	model · harness	effort	prompt	time · cost	agent activity	geometry	compare
Loading run history…

Loading run history…

[A] configuration

modeleffortharness

[B] configuration

modeleffortharness

[2] benchmark prompt

prompt

select a prompt to inspect each side’s render and 3d scene.

[3] compare configurations

compare configurations

loading model data…

[2] run

[3] provenance

[4] actions esc back to history

← history

run / —

Loading run…

[6] render preview · 2d

Preview unavailable

[7] scene model · 3d/glb

drag to orbit · scroll to zoom · r reset

[8] stats

[9] session timeline · loading trace…

loading session activity…

[3] research method

how the benchmark works

the evaluation isolates the full tool-using stack while keeping the voting experience simple and blind.

[4] pipeline

01[controlled]standardized task
every run begins with the same natural-language blender task.
02[varies]agent builds
models work through their assigned harness and reasoning setting.
03[hidden until vote]blind comparison
identities and configuration details remain hidden during voting.
04[calculated]stack-level elo
pairwise outcomes update ratings for each configuration.

[5] what is fixed · what varies

controlled: prompt, benchmark scene, comparison protocol, and vote outcomes.
what varies: model, reasoning effort, harness, and the resulting blender scene.
after voting: identities and technical run metrics are revealed without moving the renders.
low samples: new configurations remain marked provisional until enough votes accumulate.

[6] known limits

rating order: elo is replayed vote by vote in the order votes were cast, so the same votes in another order can give slightly different ratings.
shared outcomes: a tie lifts and “both bad” lowers both results at once, so these outcomes shift the overall rating level instead of trading points between the two.
blindness limits: renders are public on run pages, so a determined voter could recognise a result. identities stay hidden in the vote itself.
vote limits: each browser votes once per pair, and requests are rate-limited per network address to bound automated voting.

[2] supported agents

pi · omp · omp-mcp
omp-mcp-skill · codex · opencode

images pulled on first use

[3] contribute a run

run the benchmark locally

every run starts in a pinned docker image: one prompt per fresh container with blender and the selected agent, then the run is submitted separately.

[4] quick start

$ chmod +x bench-run
$ ./bench-run pi --model PROVIDER/MODEL:high worldtree
$ ./bench-run submit results/pi/normal/MODEL/high/worldtree/0
# existing runs: submit-run.sh --dir <path>

download bench-run requires docker · agents: pi, omp, omp-mcp, omp-mcp-skill, codex, opencode

[5] what the run does

images for pi, omp, omp-mcp, omp-mcp-skill, codex and opencode are pulled on first use.
only the selected agent’s credentials are copied into the container.
no cpu or gpu rendering is assumed; the actual blender environment is recorded with the session, preview, glb and blend file.
session logs can contain prompts, tool output and local paths — inspect them before confirming upload.

[6] after you submit

run locallybench-run executes the prompt in a fresh container
submitbench-run submit uploads the run directory
private staging queueuploads wait here, not yet public
administrator review → publishedthe run appears in history and enters voting

[3] about the benchmark

preference data for a more open 3D future.

Render Arena is an open benchmark for comparing real Blender runs across the complete agent stack. Votes are stored against an anonymous browser token so one voter cannot submit the same comparison twice. Read more →