Benchmarks
347 sources · 1,258 models × 3,038 benchmarks · 158,912 scoresFinancial Coaching Bench's Elicitation panel: the share of the scenarios' hidden fact-sheet facts (income, debts, deadlines, savings, household) that the coach asked the simulated client for, as Jev 1.13, a separate classification model, grades each fact asked for, assumed or never raised. The page gives neither the number of facts nor whether the share is pooled over facts or averaged over scenarios. Diagnostic only.
No signal or difficulty estimate. Held out of the fit by design.
| # | Model | Setting | Score | Source |
|---|---|---|---|---|
| 1 | Muse Spark 1.3 | max | 69.1 | |
| 2 | Muse Spark 1.2 | x-high | 68.2 | |
| 3 | Muse Wonder | max | 67.0 | |
| 4 | Muse Wonder Full | max | 66.0 | |
| 5 | Gemini 4 Argon | high | 63.8 | |
| 6 | Claude Sonnet 5.5 | max | 63.1 | |
| 7 | GPT-6 Astra | max | 62.9 | |
| 8 | GPT-5.6 Sol | max | 61.9 | |
| 9 | GPT-5.6 Terra | max | 61.1 | |
| 10 | GPT-6.1 Sol | max | 59.5 | |
| 11 | MiMo-V2.6-Pro | not reported | 59.1 | |
| 12 | Claude Opus 5.5 | max | 58.5 | |
| 13 | DeepSeek V4.1 Flash | high | 57.7 | |
| 14 | GPT-6 Sol | max | 55.9 | |
| 15 | GPT-5.6 Luna | max | 54.4 | |
| 16 | GLM-5.3 | max | 54.2 | |
| 17 | Gemini 3.8 Flash | high | 54.0 | |
| 18 | MiniMax M3 | — | 53.8 | |
| 19 | GPT-6 Luna | max | 53.3 | |
| 20 | Claude Fable 5.1 | max | 53.2 | |
| 21 | Grok 4.7 | x-high | 51.7 | |
| 22 | Gemini 3.5 Flash-Lite | high | 51.3 | |
| 23 | Gemini 3.1 Pro Preview | high | 49.8 | |
| 24 | Gemini 3.5 Flash | high | 49.7 | |
| 25 | Kimi K3 | max | 49.6 | |
| 26 | Gemini 3.6 Flash | high | 48.7 |
Benchmarks outside its family whose model rankings track it most reliably (Spearman ρ over at least 10 shared models).