Claude Sonnet 5.5
Key numbers
- Panel score
- 77rank 3 of 11
- Cost
- $7.28API list price
- Wall time
- 53 min
- Model time
- 39 min73% of wall time
- tok/s
- 133.0
- Tokens
- 19.5MOutput: 307k
- Cache hit
- 95%
- API calls
- 69
- Tool calls
- 76
- Renderer
- Intel UHD 630 (GPU)74 ms / frame
- H1–H6
- Pass
- Cost per point
- $0.095
How the run ended
Finished cleanly with a passing video, a procedural soundtrack and a report. Unlike the others, it kept the spec’s acceleration limits and let the top speed rise instead.
Blind panel
Scores of the three judges in every panel this video was part of.
| Panel | Letter | Anthropic | OpenAI | Median | Spread | Rank | |
|---|---|---|---|---|---|---|---|
| 14 Nov 2026 11 videos | A | 73 | 77 | 86 | 77 | 13 | 3 / 11 |
| 13 Nov 2026 10 videos | J | 75 | 76 | 79 | 76 | 4 | 3 / 10 |
| 11 Nov 2026 9 videos | I | 78 | 78 | 64 | 78 | 14 | 3 / 9 |
Sub-scores and the judges’ one-line notes from the panel of 14 Nov 2026.
| judge | Landmarks / 25 | City look / 25 | Camera & drone / 30 | Life / 10 | Overall / 10 | Total |
|---|---|---|---|---|---|---|
| Anthropic | 17 | 19 | 23 | 7 | 7 | 73 |
| OpenAI | 16 | 20 | 25 | 8 | 8 | 77 |
| 21 | 22 | 27 | 8 | 8 | 86 |
Judges’ notes
- AnthropicSolid textured city with clear drone take-off and bridge/towers, but St. Nicholas is weak and one frame hugs a wall.
- OpenAITextured streets and lively crowds support a readable chase, but the landmarks remain simplified.
- GoogleWell-paced chase flight with a clearly readable drone, textured historic streets, Charles Bridge statues and river boats, and zero geometry collisions.
Time and speed
Wall time includes rendering and encoding. Model time is the sum of all model calls.
- Model generating
- Tools & waiting
| Wall time | 53 min |
|---|---|
| Model time | 39 min |
| Tools & waiting | 14 min |
| Model share of wall time | 73% |
| tok/s | 133.0 |
| Model calls | 69 |
Tokens and cost
Tokens summed over every model response. Tokens multiplied by the provider’s price.
| Uncached input | 140 |
|---|---|
| Cache read | 18,206,332 |
| Cache write | 953,232 |
| Output | 307,153 |
| Total | 19,466,857 |
| Cache-hit ratio | 95.0% |
| Cost | $7.28API list price |
| Output tokens per point | 3,989 |
| Points per $ | 10.6 |
Tools used
Tool calls by name, as the harness logged them. Names differ between harnesses.
Rendering
| Renderer | Intel UHD 630 (GPU) |
|---|---|
| Final render | 2.1 min |
| Per frame | 74 ms |
Hard requirements H1–H6
Checked independently from the delivered MP4 with ffprobe.
| H1 | h264 High yuv420p | Pass | |
|---|---|---|---|
| H1 | faststart | Pass | |
| H2 | 1080×1920 SAR1:1 | Pass | 1080x1920 SAR 1:1 |
| H3 | 30 fps | Pass | 30 fps |
| H4 | 55-60 s | Pass | 58.000 s |
| H5 | peak≤ 20Mbps, <200MB | Pass | peak 14.72 Mbps over any 1 s, 84.6 MB |
| H6 | AAC-LC 48k stereo | Pass |
From the agent’s report
Short, verbatim quotes from the REPORT.md the agent wrote itself.
The sharp turn into the Bull Staircase (t ~ 10 s) puts the camera within ~3 m of the gatehouse face and the framing is awkward for ~1 s.
Contact sheet and route map
