SolarBench: leaderboard
SolarBench strict all-criteria macro pass rate over eight solar-portfolio operations scenarios and ten repetitions per scenario. Native sbm-1.0 payload means and source model IDs are retained; the benchmark includes cached LLM judging for writing criteria. Board: SolarBench.
Metric: Pass Rate (%). Source: solarbench.maingen.ai. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 53.75 | #16 |
| 2 | Muse Spark 1.1 | 42.5 | #52 |
| 3 | Grok 4.5 | 38.75 | #48 |
| 4 | GPT-5.6 Sol | 33.75 | #19 |
| 5 | Gemini 3.1 Pro (Preview) | 28.75 | #58 |
| 6 | Kimi K3 | 26.25 | #30 |
| 7 | GLM-5.2 | 23.75 | #73 |
| 8 | Claude Sonnet 5 | 12.5 | #57 |
| 9 | Gemini 3.5 Flash | 11.25 | #60 |
| 10 | GPT-5.6 Terra | 6.25 | #41 |
| 11 | Kimi K2.7 Code | 5 | #89 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=solarbench · How It Works · Data refreshed daily, snapshot 2026-10-20.