Gemini 4 Argon
Key numbers
- Panel score
- 52rank 8 of 10
- Cost
- $3.57API list price
- Wall time
- 37 min
- Model time
- 27 min74% of wall time
- tok/s
- 67.5
- Tokens
- 7.6MOutput: 110k
- Cache hit
- 88%
- API calls
- 77
- Tool calls
- 70
- Renderer
- Intel UHD 630 (GPU)
- H1–H6
- Pass
- Cost per point
- $0.069
How the run ended
Finished cleanly: the video passes every hard requirement, comes with a procedural soundtrack and a long report.
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 | B | 52 | 59 | 48 | 52 | 11 | 8 / 11 |
| 13 Nov 2026 10 videos | B | 54 | 55 | 65 | 55 | 11 | 6 / 10 |
| 11 Nov 2026 9 videos | C | 57 | 52 | 55 | 55 | 5 | 6 / 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 | 14 | 11 | 15 | 7 | 5 | 52 |
| OpenAI | 13 | 12 | 23 | 6 | 5 | 59 |
| 14 | 10 | 16 | 4 | 4 | 48 |
Judges’ notes
- AnthropicFlat, washed-out white/grey look; dome and towers present but generic, with one frame stuck inside an arcade.
- OpenAIThe drone and route are mostly readable, but flat surfaces and repetitive buildings weaken the city.
- GoogleDrone with bright LEDs is visible, but the city suffers from flat grey ground slabs, missing blue river water, and washed-out finale lighting.
Time and speed
Wall time includes rendering and encoding. Model time is the sum of all model calls.
- Model generating
- Tools & waiting
| Wall time | 37 min |
|---|---|
| Model time | 27 min |
| Tools & waiting | 9.8 min |
| Model share of wall time | 74% |
| tok/s | 67.5 |
| Model calls | 77 |
Tokens and cost
Tokens summed over every model response. Tokens multiplied by the provider’s price.
| Uncached input | 902,669 |
|---|---|
| Cache read | 6,591,807 |
| Cache write | 0 |
| Output | 110,302 |
| Total | 7,604,778 |
| Cache-hit ratio | 88.0% |
| Cost | $3.57API list price |
| Output tokens per point | 2,121 |
| Points per $ | 14.6 |
Tools used
Tool calls by name, as the harness logged them. Names differ between harnesses.
Rendering
| Renderer | Intel UHD 630 (GPU) |
|---|---|
| Final render | – |
| Per frame | – |
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 unset (= square) |
| H3 | 30 fps | Pass | 30 fps |
| H4 | 55-60 s | Pass | 58.000 s |
| H5 | peak≤ 20Mbps, <200MB | Pass | peak 18.27 Mbps over any 1 s, 84.7 MB |
| H6 | AAC-LC 48k stereo | Pass |
From the agent’s report
Short, verbatim quotes from the REPORT.md the agent wrote itself.
3 minor clearance adjustments were applied and smoothed, ensuring 0 frames clip through any wall, roof, statue, tree, or terrain.
Contact sheet and route map
