DecisionPivot - Core Pair Accuracy: leaderboard
DecisionPivot v1.1:7800coreitems,13generated task families,macro-averaged accuracy and contrast-group pair accuracy;40decision models and14general controls. Exact revisions and native inference in AppendixB1; BF16 except BlinkFP8 andd1-omniFP16. Input-limit failures are incorrect. Hosted Jev1.13 is distinct from JEV students. Reference35-task score is chance-corrected/floored and layer-averaged. Score (%).
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around January 2027. 54 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 (Low) | 82.2 | #37 (GLM-5.3) |
| 2 | Gemma 4 31B (IT) (Non-reasoning) | 37.2 | #177 (Gemma 4 31B (IT)) |
| 3 | Jev 1.13 | 31 | #380 |
| 4 | DeepSeek V4 Pro (Non-reasoning) | 14.3 | #106 (DeepSeek V4 Pro) |
| 5 | DeepSeek V4.1 Flash (Non-reasoning) | 14.2 | #35 (DeepSeek V4.1 Flash) |
| 6 | Qwen 3.8 27B (Non-reasoning) | 12.9 | #111 (Qwen 3.8 27B) |
| 7 | Qwen 3.5 9B (Non-reasoning) | 11.4 | #383 (Qwen 3.5 9B) |
| 8 | Qwen 3.6 35B A3B (Non-reasoning) | 9.9 | #229 (Qwen 3.6 35B A3B) |
| 9 | Qwen 3.5 4B (Non-reasoning) | 9.6 | #485 (Qwen 3.5 4B) |
| 10 | Gemma 3 12B (IT) | 8.5 | #667 |
| 11 | Phi-4 | 8.2 | #706 |
| 12 | Granite 3.3 8B Instruct | 4.8 | #815 |
| 13 | Qwen 3.5 2B (Non-reasoning) | 3.6 | #1149 (Qwen 3.5 2B) |
| 14 | Qwen 3.5 0.8B (Non-reasoning) | 1.5 | #1430 (Qwen 3.5 0.8B) |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=decisionpivot-core-pair-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-22.