Kvark Sparkle Release - GPQA Diamond 198x5: leaderboard
Official KvarkSparkle1.0release,GPQA Diamond 198x5,nativeaccuracypercent. BothmodelsFP8weights,thinkingon,64ktokencap,sameprompts,temp1,top_p0.95,top_k20. Exactitem/seedcohorts retained;not fullLCBv6 orfullMMLUPro. Pairedbootstrapintervals,medianreasoningtokens,latency andcaphits are diagnosticcolumns,not scores. Sparkle isdistincttrainedmerge;FP8artifactlink verified.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3.8-27B (FP8) (Thinking) | 89.4 | |
| 2 | Kvark-Sparkle-1.0-27B-FP8 (Thinking) | 89.3 |
Interactive version: theaggregate.ai/benchmark?slug=kvark-sparkle-release-gpqa-diamond-198x5 · How It Works · Data refreshed daily, snapshot 2026-11-12.