prague-bench

← All runs

Qwen3.8 Max

FinishedAlibabaOpenCode · OpenRouterEffort: medium13 Nov 2026spec 666c07f57d6e

Video preview · 540 × 960 · 5.8 MB

Key numbers

Panel score
63rank 4 of 11
Cost
$18.74 †amount billed
Wall time
4 h 6 min
Model time
3 h 59 min97% of wall time
tok/s
30.9
Tokens
69.2MOutput: 443k
Cache hit
99%
API calls
197
Tool calls
197
Renderer
Intel UHD 630 (GPU)115 ms / frame
H1–H6
Pass
Cost per point
$0.298

How the run ended

Delivered the video, contact sheet, route map and report, then kept polishing the camera until OpenRouter refused the next request for lack of credit. The files on disk were its last complete version and were scored as they are.

Blind panel

Scores of the three judges in every panel this video was part of.

Blind panel
PanelLetterAnthropicOpenAIGoogleMedianSpreadRank
14 Nov 2026 11 videosF63537863254 / 11
13 Nov 2026not in this panel
11 Nov 2026not in this panel

Sub-scores and the judges’ one-line notes from the panel of 14 Nov 2026.

Sub-scores and the judges’ one-line notes from the panel of 14 Nov 2026.
judgeLandmarks / 25City look / 25Camera & drone / 30Life / 10Overall / 10Total
Anthropic1517187663
OpenAI1014186553
Google2020247778

Judges’ notes

  • AnthropicTextured and colorful with boats and a good finale, but early frames are squeezed between walls and some views are skewed.
  • OpenAIColorful textured streets help, but malformed geometry, awkward framing and crude landmarks disrupt the flight.
  • GoogleCrisp architectural detail on St. Vitus, Mostecka street, and the bridge towers, though the Vltava river looks overly dark and the drone is smaller in frame.

Time and speed

Wall time includes rendering and encoding. Model time is the sum of all model calls.

  • Model generating
  • Tools & waiting
  1. Wall time4 h 6 min97% model
Time and speed
Wall time4 h 6 min
Model time3 h 59 min
Tools & waiting7.0 min
Model share of wall time97%
tok/s30.9
Model calls197

Tokens and cost

Tokens summed over every model response. Tokens multiplied by the provider’s price.

Tokens and cost
Uncached input439,668
Cache read67,866,209
Cache write491,843
Output443,250of which 221,822 reasoning
Total69,240,970
Cache-hit ratio98.7%
Cost$18.74 †amount billedOpenCode’s own estimate: $18.83
Output tokens per point7,036
Points per $3.4

Tools used

Tool calls by name, as the harness logged them. Names differ between harnesses.

  1. shell85
  2. edit48
  3. read41
  4. write23

Rendering

Rendering
RendererIntel UHD 630 (GPU)
Final render1.8 min
Per frame115 ms

Hard requirements H1–H6

Checked independently from the delivered MP4 with ffprobe.

Hard requirements H1–H6
H1h264 High yuv420pPass
H1faststartPass
H21080×1920 SAR1:1Pass1080x1920 SAR 1:1
H330 fpsPass30 fps
H455-60 sPass58.000 s
H5peak≤ 20Mbps, <200MBPasspeak 14.13 Mbps over any 1 s, 76.0 MB
H6AAC-LC 48k stereoPass

From the agent’s report

Short, verbatim quotes from the REPORT.md the agent wrote itself.

The Bull Staircase passage (t ≈ 9–12 s) is the weakest shot: the sloped side walls read as stacked masonry and the moment is brief/dark.

Contact sheet and route map