Gemini 4 Hydrogen
Key numbers
- Panel score
- 48rank 9 of 11
- Cost
- $3.01API list price
- Wall time
- 20 min
- Model time
- 14 min71% of wall time
- tok/s
- 190.6
- Tokens
- 6.9MOutput: 162k
- Cache hit
- 87%
- API calls
- 67
- Tool calls
- ~63
- Renderer
- Intel UHD 630 (GPU)56 ms / frame
- H1–H6
- Pass
- Cost per point
- $0.063
How the run ended
The video is finished and passes every hard requirement, but Antigravity stopped the turn with a provider error right after the final encode. The agent never ran its own checks or wrote REPORT.md. The blind panel judges only the video, so the score is unaffected.
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 | E | 41 | 48 | 52 | 48 | 11 | 9 / 11 |
| 13 Nov 2026 10 videos | H | 57 | 45 | 37 | 45 | 20 | 8 / 10 |
| 11 Nov 2026 | not in this panel | ||||||
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 | 12 | 10 | 10 | 6 | 3 | 41 |
| OpenAI | 11 | 10 | 17 | 6 | 4 | 48 |
| 15 | 12 | 15 | 5 | 5 | 52 |
Judges’ notes
- AnthropicSeveral frames pressed against or under geometry, tilted buildings, flat washed look and a hard-to-read drone.
- OpenAIRecognisable route elements are undermined by flat plazas, crude structures and intrusive foreground geometry.
- GoogleFeatures a green St. Nicholas dome and river views, but looks hazy and washed out with a flat wall slab at 0 s and an enclosed dark passage at 36.2 s.
Time and speed
Wall time includes rendering and encoding. Model time is the sum of all model calls.
- Model generating
- Tools & waiting
| Wall time | 20 min |
|---|---|
| Model time | 14 min |
| Tools & waiting | 5.8 min |
| Model share of wall time | 71% |
| tok/s | 190.6 |
| Model calls | 67 |
Tokens and cost
Tokens summed over every model response. Tokens multiplied by the provider’s price.
| Uncached input | 903,036 |
|---|---|
| Cache read | 5,854,966 |
| Cache write | 0 |
| Output | 162,166 |
| Total | 6,920,168 |
| Cache-hit ratio | 86.6% |
| Cost | $3.01API list price |
| Output tokens per point | 3,378 |
| Points per $ | 16.0 |
Tools used
Tool calls by name, as the harness logged them. Names differ between harnesses. Approximate counts from the run log.
Rendering
| Renderer | Intel UHD 630 (GPU) |
|---|---|
| Final render | 1.6 min |
| Per frame | 56 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 15.06 Mbps over any 1 s, 68.7 MB |
| H6 | AAC-LC 48k stereo | Pass |
From the agent’s report
The agent did not write a REPORT.md.
Contact sheet and route map
