Reific Environments - Science and Compute: leaderboard
Reific's Science and Compute environments evaluate agents on bioreactors, GPU clusters and storage fleets. Records arrive over time; agents read, run code, ask people and act with tools. The score is mean graded reward across environments in the family, not a task pass rate. The shared source protocol uses high reasoning and fresh repeated trials; numeric checks are deterministic, other outputs average two judges. Exact endpoints, numeric budgets and judge identities remain source-scoped and unreported.
Metric: Mean graded reward (%). Source: reific.co. Saturation forecast: Rough model projection: around 2026. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Reific Run: native=Claude%20Opus%205.5; cfg=ba20783f598bfef7) | 29.7 | |
| 2 | GPT-6 Astra (Reific Run: native=GPT-6%20Astra; cfg=ae3b2e98cfafdb96) | 22.2 | |
| 3 | GPT-6.1 Sol (Reific Run: native=GPT-6.1%20Sol; cfg=a8a089cd5143b78d) | 16.7 | |
| 4 | GPT-6.1 Terra (Reific Run: native=GPT-6.1%20Terra; cfg=bb6065933bf837d0) | 16.7 | |
| 5 | GPT-6 Luna (Reific Run: native=GPT-6%20Luna; cfg=889ff97516aff056) | 16.7 | |
| 6 | Grok 4.7 (Reific Run: native=Grok%204.7; cfg=5e1476703201cde7) | 15.6 | |
| 7 | Claude Sonnet 5.5 (Reific Run: native=Claude%20Sonnet%205.5; cfg=ccd0a9d2ec8c2b92) | 15 | |
| 8 | Claude Haiku 5.5 (Reific Run: native=Claude%20Haiku%205.5; cfg=ccd7083e7cd1893b) | 14 | |
| 9 | Muse Wonder (Reific Run: native=Muse%20Wonder; cfg=c8cbdcd9fab6d787) | 13.9 | |
| 10 | Gemini 4 Argon (Reific Run: native=Gemini%204%20Argon; cfg=894138d508356a5e) | 13.3 |
Interactive version: theaggregate.ai/benchmark?slug=reific-environments-science-and-compute · How It Works · Data refreshed daily, snapshot 2026-11-05.