Ripple 65B Preview Release - Instruct - SciCode (Inspect AI): leaderboard

Ricursive Ripple-65B Preview native instruct comparison, immutable card revision e27e171792f303d74c048184e8259ba663cfa0d5. Full BF16 precision and maximum effort are applied only to Ricursive in-house evaluations and its own instruct column; imported official-card cells retain unreported precision/effort and distinct source-qualified identities. Native percentages, higher is better. Ripple-65B Preview:in-house/Inspect AI; Nemotron 3 Ultra 550B:official-card/Inspect AI; Inkling 975B:official-card/Inspect AI; Muse Spark Open 816B:official-card/Inspect AI; Reflection Beam 501B:official-card/Inspect AI. Terminal-Bench self-reported and Terminus2 comparisons are separate; native table80.5 is retained, not replaced by prose80.52. No throughput/cost/model-size row is a benchmark. This source family alone does not establish independent main-rank eligibility.

Metric: Native SciCode (Inspect AI) (%). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 5 models tracked.

Top models

#ModelScoreOverall rank
1Muse Spark Open 816B (Ripple Release Run: phase=Instruct;origin=official-card;precision=unreported;effort=unreported;harness=Inspect AI)58.5
2Reflection Beam 501B (Ripple Release Run: phase=Instruct;origin=official-card;precision=unreported;effort=unreported;harness=Inspect AI)49.2
3Inkling 975B (Ripple Release Run: phase=Instruct;origin=official-card;precision=unreported;effort=unreported;harness=Inspect AI)46.1
4Nemotron 3 Ultra 550B (Ripple Release Run: phase=Instruct;origin=official-card;precision=unreported;effort=unreported;harness=Inspect AI)44.6
5Ripple-65B Preview (Ripple Release Run: phase=Instruct;origin=in-house;precision=BF16;effort=maximum;harness=Inspect AI)39.4

Interactive version: theaggregate.ai/benchmark?slug=ripple-65b-preview-release-instruct-scicode-inspect-ai · How It Works · Data refreshed daily, snapshot 2026-11-06.