prague-bench

← All runs

Claude Opus 5.5 (run 2, older spec)

FinishedAnthropicClaude CodeEffort: default10 Nov 2026spec ed8ccc175f90 (older spec)

Video preview · 540 × 960 · 5.7 MB

Key numbers

Panel score
–not scored
Cost
$24.21API list price
Wall time
2 h
Model time
47 min40% of wall time
tok/s
98.0
Tokens
26.5MOutput: 278k
Cache hit
89%
API calls
89
Tool calls
89
Renderer
SwiftShader (software, CPU)2,100 ms / frame
H1–H6
Pass

How the run ended

Finished with a passing video, but on an older version of the spec and with a software (SwiftShader) render that alone took over an hour. Its time and cost are not comparable with the later runs, so it is kept for reference and was not sent to the blind panel.

Provisional score (orchestrator, superseded): 83

Time and speed

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

  • Model generating
  • Tools & waiting
  1. Wall time2 h40% model
Time and speed
Wall time2 h
Model time47 min
Tools & waiting1 h 12 min
Model share of wall time40%
tok/s98.0
Model calls89

Tokens and cost

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

Tokens and cost
Uncached input186
Cache read23,460,960
Cache write2,791,355
Output278,115
Total26,530,616
Cache-hit ratio89.4%
Cost$24.21API list price

Tools used

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

  1. Bash59
  2. Write17
  3. Read10
  4. Monitor2
  5. ToolSearch1

Rendering

Rendering
RendererSwiftShader (software, CPU)
Final render1 h 2 min
Per frame2,100 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 15.79 Mbps over any 1 s, 80.8 MB
H6AAC-LC 48k stereoPass

From the agent’s report

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

With the specified limits of 12 and 7 m/s², the fastest the route's many tight corners allow is about 58.6 s. That is more than the time available.

Contact sheet and route map