prague-bench

← All runs

Grok 4.7

FinishedxAICursor · CursorEffort: medium11 Nov 2026spec 666c07f57d6e

Video preview · 540 × 960 · 5.0 MB

Key numbers

Panel score
30rank 10 of 11
Cost
n/anot recorded
Wall time
31 min
Model time
–
tok/s
–
Tokens
–
Cache hit
–
API calls
–
Tool calls
122
Renderer
Intel UHD 630 (GPU)~80 ms / frame
H1–H6
Pass

How the run ended

Finished with a passing video and a report. Cursor does not store token usage locally, so tokens, cost and generation speed are unknown.

Blind panel

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

Blind panel
PanelLetterAnthropicOpenAIGoogleMedianSpreadRank
14 Nov 2026 11 videosH37293030810 / 11
13 Nov 2026 10 videosG44324544139 / 10
11 Nov 2026 9 videosE3837433868 / 9

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
Anthropic10996337
OpenAI47115229
Google10873230

Judges’ notes

  • AnthropicGrid-window slab buildings, camera repeatedly jammed against facades and dark blocks; only the bridge reads well.
  • OpenAIRepeated facade collisions, broken-looking passages and blocklike landmarks severely weaken the otherwise populated route.
  • GoogleDark, muddy palette with severe camera collisions where walls and dark geometry completely block the view at 7.3 s, 10.9 s, and 36.2 s.

Time and speed

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

Time and speed
Wall time31 min
Model time–
Tools & waiting–
Model share of wall time–
tok/s–
Model calls–

Tokens and cost

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

No data.

Tools used

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

  1. edit43
  2. read41
  3. shell32
  4. grep5
  5. glob1

Rendering

Rendering
RendererIntel UHD 630 (GPU)
Final render1.3 min
Per frame80 ms (approximate)

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 14.17 Mbps over any 1 s, 61.9 MB
H6AAC-LC 48k stereoPass

From the agent’s report

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

On the New Castle Steps the chase looks through a gap between walls. A pedestrian on the stair can read as standing on a roof below.

Contact sheet and route map