Gemini 3.8 Flash (Google)
google/gemini-3.8-flash · Rank 6 of 6
Taste Elo1384.0
Duels (W–L–T)28 0–10–18
Slop density18.7%
Artifacts judged8
Rubric profile
Mean score per dimension on the 0–4 level scale, with bootstrap confidence interval and sample size. Dimensions are omitted when no artifact qualifies (Communication without an explanation, Test taste when tests are out of scope); the composite renormalizes.
| Dimension | Mean | 95% CI | n |
| Clarity & structure |
3.17 |
3.09–3.25 |
7 |
| Idiom & economy |
2.89 |
2.70–3.08 |
7 |
| Signal discipline |
2.65 |
2.27–3.07 |
7 |
| Communication |
2.40 |
1.61–2.89 |
7 |
| Test taste |
3.10 |
2.89–3.24 |
3 |
Slop diagnostics
Whole-run slop density: 18.7% (0.187 raw), the mean of twelve binary diagnostics across this model's artifacts. Per-diagnostic detail appears here once raw judge output for this model's demo items is published under results/raw/.
Harness provenance
| Harness | openrouter-chat-completions openrouter-chat-2026-09-22 |
| Temperature | 0.3 |
| Generated | 2026-09-22T14:11:46.189Z – 2026-09-22T14:27:59.127Z |
| Source | generated |
| Published demo items | 4 |
Judge
Generation jev-1.13, frozen for this run. Serving snapshot ids recorded per run:
| typesafe/jev-1.13-20260917 | 204 |
How these numbers are produced is documented on the Methodology page.