~ same prompt, different LLM — every model gets the same brief in Blender, you judge the render ~
OpenAI / GPT 6 Astra / thinking:max has entered the arena NEW!
Render Window — hero
hero render
camera + framing set up by the AI contestant
3D Viewport — goldfish User Persp
LOADING 3D VIEW
$0.00Cost
785sDuration
22Tool calls
22Turns
Blender 5.1.0MCP 1.0.02026-08-25
Tokens: 271571/10527Vision: Votes W-L-T: 0-0-0reasoning: effort lowNO DONE
B-BENCH — TRANSCRIPT Ox Alpha / thinking:low · r1 · 2026-08-25
RUN LOG — goldfish · Ox Alpha / thinking:low · r1 · model_error
SYSTEM PROMPTthe control — every model on this task got these instructions
PROMPTblender-bench v8HARNESSraw-v3
SHOW FULL PROMPT · 3364 chars
You are participating in Blender-Bench, a public benchmark where AI models complete the same Blender task.

Your result will be compared head-to-head with another model. Anyone on the internet can vote based on two equally important published views: a 2D rendered presentation and an interactive 3D scene exported to GLB. Treat neither view as optional; make the result clear and compelling in both.

Complete the request as well as you can. The benchmark does not prescribe a workflow or visual style; use whichever Blender techniques you consider appropriate. Make the important requirements evident in the published result, save a complete scene, and present the work clearly in every requested output.

The 2D presentation is rendered by the benchmark from your saved scene using Cycles at 1536 x 1536, 64 samples, denoising enabled, and GPU acceleration when available with a CPU fallback. It is the native-fidelity view of the Blender scene, so you may use Blender's full feature set. Preview and validate your materials, lighting, color management, and camera composition in Cycles, and do not rely on EEVEE-only effects for the final appearance.

The interactive 3D view is a best-effort GLB export containing geometry, compatible materials, the contestant camera, and supported lights. Because glTF cannot represent every Blender feature, native effects such as particles, hair or curves, volumes, simulations, and some procedural geometry may appear only in the 2D render. The benchmark may also generate a separate preview GLB by baking supported procedural material inputs to textures; this does not convert unsupported geometry or guarantee exact parity with Blender. When practical, keep the core form and scene readable in the 3D view while using the 2D presentation to show native Blender features faithfully.

## Harness Instructions

You are operating Blender through the provided tools. Tool calls execute one at a time, in the order you issue them, against a single live Blender instance — each call sees the state left by the previous one, and issuing several calls in one message gains no parallelism. The scene starts empty. Build what the user's prompt asks for in the current scene. Work step by step and verify your own work.

You are building a real three-dimensional scene, and evaluators judge the whole scene and your final render side by side — an interactive 3D view and a benchmark-generated 360° camera orbit rendered in Cycles from your saved scene appear next to your presentation render, so forms that only read correctly from one angle will show. Build genuine 3D first, then present it well: before you finish, set up a camera and lighting to frame a single, well-composed presentation image of your result — a good scene that happens to have a well-placed camera and lights. The final benchmark capture overrides the scene render settings and renders in Cycles at 1536x1536, 64 samples, with denoising and GPU acceleration (CPU fallback). Preview and validate materials, lighting, color management, and composition in Cycles; do not rely on EEVEE-only effects. Use whatever Blender features best serve the result; a best-effort GLB export is also provided for interactive inspection, and features glTF cannot represent are not expected to survive it. When you are finished, reply with the single word DONE and no tool calls.
--TASK goldfish: ## Task Complete the following Blender task: <task> A single goldfish as the hero of the shot: rounded body, a big flowing tail fin, dorsal and side fins, and an eye on each side, in an unmistakable goldfish orange. The floor is a shape any child would name instantly. The ceiling is life: a gentle S-curve in the body as if mid-swim, translucency toward the fin edges, a subtle scale pattern, an…
46:49ai>I'll build a goldfish hero shot: procedurally modeled S-curved body, flowing tail, translucent fins, scale-pattern material, circular "instantly nameable" floor, and a watery teal backdrop.
46:49>>execute_blender_code({"code":"\nimport bpy, math\nfrom math import sin, pi\n\n# clean scene\nbpy.ops.object.select_all(action='SELECT')\nbpy.ops.object.delete()\nfor m in list(bpy.data.meshes):\n bpy.data.meshes.remove(m)\n\nL = 3.0 # body length\ndef sdx(y): # S-curve offset in X\n return 0.32*sin((y+1.5)*(2*pi/L))\n\n# ---------- BODY ----------\nbpy.ops.mesh.primitive_uv_sphere_add(segments=48, …)
46:50<<{ "status": "error", "message": "Traceback (most recent call last):\n File \"<repo>\\.runner\\blender-mcp-v1.0.0\\addon\\mcp_to_blender_server.py\", line 230, in _execute_code\n exec(code, namespace)\n ~~~~^^^^^^^^^^^^^^^^^\n File \"<string>\", line 48, in <module>\nAttributeError: module 'bpy' has no attribute 'objects'\n" }
47:07>>execute_blender_code({"code":"\nimport bpy, math\nfrom math import sin, pi, sqrt\n\nbpy.ops.object.select_all(action='SELECT'); bpy.ops.object.delete()\n\nL = 3.0\ndef sdx(y): return 0.32*sin((y+1.5)*(2*pi/L))\n\n# BODY\nbpy.ops.mesh.primitive_uv_sphere_add(segments=48, ring_count=32)\nbody = bpy.context.active_object\nbody.name = \"Goldfish\"\nbody.scale = (0.55, 1.5, 0.78)\nbpy.ops.object.transform_apply(scale=True)…)
47:08<<{ "status": "ok", "result": { "objects": [ "DorsalFin", "Eye.L", "Eye.R", "Goldfish", "PectoralFin.L", "PectoralFin.R", "TailFin" ] } }
47:32ai>Now materials, floor, backdrop, lights, camera.
47:32>>execute_blender_code({"code":"\nimport bpy\n\ndef newmat(name):\n m=bpy.data.materials.new(name); m.use_nodes=True\n return m, m.node_tree.nodes[\"Principled BSDF\"], m.node_tree\n\n# ---- BODY: goldfish orange with scale pattern (procedural) ----\nmb,bsdf,nt=newmat(\"GoldfishSkin\")\nout=nt.nodes[\"Material Output\"]\ntex=nt.nodes.new(\"ShaderNodeTexCoord\")\nmapping=nt.nodes.new(\"ShaderNodeMapping\"); mapping…)
47:34<<{ "status": "error", "message": "Traceback (most recent call last):\n File \"<repo>\\.runner\\blender-mcp-v1.0.0\\addon\\mcp_to_blender_server.py\", line 230, in _execute_code\n exec(code, namespace)\n ~~~~^^^^^^^^^^^^^^^^^\n File \"<string>\", line 48, in <module>\nNameError: name 'tail' is not defined\n" }
47:55>>execute_blender_code({"code":"\nimport bpy\ntail=bpy.data.objects[\"TailFin\"]; dorsal=bpy.data.objects[\"DorsalFin\"]\n\ndef newmat(name):\n m=bpy.data.materials.new(name); m.use_nodes=True\n return m, m.node_tree.nodes[\"Principled BSDF\"], m.node_tree\n\n# BODY skin\nmb,bsdf,nt=newmat(\"GoldfishSkin\")\ntex=nt.nodes.new(\"ShaderNodeTexCoord\")\nmapping=nt.nodes.new(\"ShaderNodeMapping\"); mapping.inputs['Scal…)
47:56<<{ "status": "ok", "result": { "ok": true } }
48:11>>execute_blender_code({"code":"\nimport bpy, math\nfrom math import radians\n# Camera: three-quarter view from front-left, slightly below eye level of fish\ncam_data=bpy.data.cameras.new(\"Cam\"); cam=bpy.data.objects.new(\"Camera\",cam_data)\nbpy.context.scene.collection.objects.link(cam)\ncam.location=(6.2,-5.5,1.4)\n# point at fish center (0,0.8,0)\nimport mathutils\ntarget=mathutils.Vector((0,0.9,0))\nd=target-math…)
48:12<<{ "status": "ok", "result": { "ok": true }, "stdout": "Info: Saved as \"goldfish.blend\"\n" }
48:19>>render_thumbnail_to_path({"output_path":"<repo>/thumb.png"})
49:13<<{ "status": "ok", "result": { "status": "ok", "filepath": "<home>\\AppData\\Local\\Temp\\blender_a42196\\blender_mcp\\thumb.png" } }
49:20>>execute_blender_code({"code":"\nimport bpy\nimg = None\n# find the just-rendered image\nfor i in reversed(bpy.data.images):\n if i.name.lower().startswith('render'): img=i; break\nprint([i.name for i in bpy.data.images])\n"})
49:21<<{ "status": "ok", "result": {}, "stdout": "['Render Result', 'Viewer Node']\n" }
49:29>>get_screenshot_of_area_as_image({"area_ui_type":"VIEW_3D"})
49:31<<
49:31imgimage delivered — get_screenshot_of_area_as_image
50:05>>render_viewport_to_path({"output_path":"<repo>/preview.png"})
50:57<<{ "status": "ok", "result": { "status": "ok", "filepath": "<home>\\AppData\\Local\\Temp\\blender_a42196\\blender_mcp\\preview.png" } }
51:02>>get_screenshot_of_area_as_image({"area_ui_type":"IMAGE_EDITOR"})
51:02<<Error executing tool get_screenshot_of_area_as_image: No area with type 'IMAGE_EDITOR' found. Available: OUTLINER, PROPERTIES, TIMELINE, VIEW_3D
51:34>>execute_blender_code({"code":"\nimport bpy, os\n# open the render preview in image editor by loading it as image and setting an area? Simpler: report file\nresult={\"path\":\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp/preview.png\"}\n"})
51:34<<{ "status": "ok", "result": { "path": "<home>/AppData/Local/Temp/blender_a42196/blender_mcp/preview.png" } }
51:59>>execute_blender_code({"code":"\nimport bpy\n# load preview into an image editor area for screenshot\nimg=bpy.data.images.load(\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp/preview.png\", check_existing=True)\nfor a in bpy.context.screen.areas:\n if a.type=='VIEW_3D':\n # split to create image editor\n override={'area':a}\n bpy.ops.screen.area_split(override, direction='VERTICAL', facto…)
52:00<<{ "status": "error", "message": "Traceback (most recent call last):\n File \"<repo>\\.runner\\blender-mcp-v1.0.0\\addon\\mcp_to_blender_server.py\", line 230, in _execute_code\n exec(code, namespace)\n ~~~~^^^^^^^^^^^^^^^^^\n File \"<string>\", line 9, in <module>\nTypeError: _bpy.ops.call() argument 3 must be str, not dict\n" }
52:06>>execute_blender_code({"code":"\nimport bpy\nimg=bpy.data.images.load(\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp/preview.png\", check_existing=True)\na=[a for a in bpy.context.screen.areas if a.type=='VIEW_3D'][0]\nbpy.ops.screen.area_split(direction='VERTICAL', factor=0.5)\nfor a2 in bpy.context.screen.areas:\n if a2.type=='VIEW_3D' and a2!=a:\n a2.type='IMAGE_EDITOR'\n a2.spaces.active.im…)
52:07<<{ "status": "ok", "result": { "ok": true } }
52:20>>get_screenshot_of_area_as_image({"area_ui_type":"IMAGE_EDITOR"})
52:23<<Error executing tool get_screenshot_of_area_as_image: No area with type 'IMAGE_EDITOR' found. Available: OUTLINER, PROPERTIES, TIMELINE, VIEW_3D
53:03>>execute_blender_code({"code":"\nimport bpy\ninfo=[(a.type, a.as_pointer()) for a in bpy.context.screen.areas]\nfor a in bpy.context.screen.areas:\n if a.type=='VIEW_3D':\n a.type='IMAGE_EDITOR'\n try: a.spaces.active.image=bpy.data.images[\"preview.png\"] or None\n except Exception as e: print(e)\n break\nresult=[a.type for a in bpy.context.screen.areas]\n"})
53:04<<{ "status": "error", "message": "The `result` variable must be a dict, not list. Wrap your return value: `result = {\"key\": value}`" }
53:29>>get_screenshot_of_area_as_image({"area_ui_type":"IMAGE_EDITOR"})
53:32<<
53:32imgimage delivered — get_screenshot_of_area_as_image
54:25ai>The body material washed out — the wave texture multiplied over the base color. I'll rebuild the skin: orange gradient (deep orange back → pale belly), wave only as a bump for the scale pattern, and make fins more saturated.
54:25>>execute_blender_code({"code":"\nimport bpy\nmat=bpy.data.materials[\"GoldfishSkin\"]; nt=mat.node_tree\nnt.nodes.clear()\nout=nt.nodes.new(\"ShaderNodeOutputMaterial\")\nbsdf=nt.nodes.new(\"ShaderNodeBsdfPrincipled\")\ntex=nt.nodes.new(\"ShaderNodeTexCoord\")\n\n# color: gradient along Z (belly->back)\ngmap=nt.nodes.new(\"ShaderNodeMapping\"); gmap.inputs['Location'].default_value=(0,0,0.78)\ngmap.inputs['Scale'].defa…)
54:25<<{ "status": "ok", "result": { "ok": true }, "stdout": "Info: Saved as \"goldfish.blend\"\n" }
54:32>>render_viewport_to_path({"output_path":"<repo>/preview2.png"})
55:24<<{ "status": "ok", "result": { "status": "ok", "filepath": "<home>\\AppData\\Local\\Temp\\blender_a42196\\blender_mcp\\preview2.png" } }
55:34>>execute_blender_code({"code":"\nimport bpy\nimg=bpy.data.images.load(\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp/preview2.png\", check_existing=True)\nfor a in bpy.context.screen.areas:\n if a.type=='IMAGE_EDITOR':\n a.spaces.active.image=img\nresult={\"ok\":True}\n"})
55:35<<{ "status": "ok", "result": { "ok": true } }
55:38>>get_screenshot_of_area_as_image({"area_ui_type":"IMAGE_EDITOR"})
55:38<<
55:38imgimage delivered — get_screenshot_of_area_as_image
55:57>>execute_blender_code({"code":"\nimport bpy\nimg=bpy.data.images[\"preview2.png\"]\nw,h=img.size\npx=list(img.pixels)\n# sample some pixels\nn=len(px)\nstep=w*h//10\nsamples=[(round(px[i*4],2),round(px[i*4+1],2),round(px[i*4+2],2)) for i in range(0,10)]\n# also center pixel\ncx=(h//2*w + w//2)*4\nresult={\"samples\":samples,\"center\":(round(px[cx],2),round(px[cx+1],2),round(px[cx+2],2))}\n"})
55:57<<{ "status": "ok", "result": { "samples": [ [ 0.57, 0.64, 0.65 ], [ 0.56, 0.65, 0.66 ], [ 0.55, 0.64, 0.66 ], [ 0.55, 0.64, 0.66 ], [ 0.55, 0.64, 0.66 ], [ 0.56, 0.65, 0.66 ], …
56:54>>execute_blender_code({"code":"\nimport bpy, os\nb=bpy.data.objects[\"Goldfish\"]\nresult={\"slots\":[(s.name, s.material.name if s.material else None) for s in b.material_slots],\n \"files\":{f: os.path.getsize(\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp/\"+f) for f in os.listdir(\"<home>/AppData/Local/Temp/blender_a42196/blender_mcp\")}}\n"})
56:54<<{ "status": "ok", "result": { "slots": [ [ "GoldfishSkin.001", "GoldfishSkin.001" ] ], "files": { "preview.png": 2591334, "preview2.png": 2591334, "thumb.png": 2591334 } } }
57:21--status: model_error
>
FULL CALL INSPECTOR
OPEN RAW JSONL ↗
0 SHOWN
LOADING COMPLETE TRANSCRIPT…
Votes cast on this bench: 000482 · 157 runs · 25 contestants
Best viewed with Netscape Navigator 4.07 at 800×600 · real runs, live ratings
Blender-Bench — which robot blends best?