LifeSci Diamond (2026-11) - Arca Workflow - Partial Credit: leaderboard
The November 3 LifeSci Diamond cohort has 55 regulatory affairs tasks, one saved answer per task, a common rubric and an LLM jury (Sonnet 5, GPT-5.6 Sol and Gemini 3.1 Pro Preview). Results combine earlier batches with six November additions. The exact scored task/rubric manifest is not public; 11 keys lack expert review. Arca’s chat workflow uses its own tools and system instructions through OpenRouter, with a 20-minute limit per turn. Three fixed API configurations and Auto are separate runs; saved Auto answers selected GPT-6.1 Terra on all 55 tasks in this edition. Effective provider settings were not recorded. The published mean gives partial credit and is distinct from strict pass rate. Native rounded means are retained.
Metric: Mean partial-credit score (0-100). Source: www.arca.inc. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Arca Run: LifeSci Diamond 2026-11; Fixed) | 86.6 | |
| 2 | Claude Opus 5.5 (Arca Run: LifeSci Diamond 2026-11; Fixed) | 83.3 | |
| 3 | Claude Opus 5 (Arca Run: LifeSci Diamond 2026-11; Fixed) | 82.4 | |
| 4 | GPT-6.1 Terra (Arca Run: LifeSci Diamond 2026-11; Auto) | 82.1 |
Interactive version: theaggregate.ai/benchmark?slug=lifesci-diamond-2026-11-arca-workflow-partial-credit · How It Works · Data refreshed daily, snapshot 2026-11-05.