MixtureofMerges-MoE-4x7B-v10-mixtralv0.3: benchmark results
jsfs11 release with a Mixtral-architecture configuration: 4 local experts, 2 selected per token and 32 decoder layers. The hidden-gate merge recipe uses mistralai/Mistral-7B-Instruct-v0.3, Kukedlc/NeuralSynthesis-7B-v0.1, mlabonne/AlphaMonarch-7B, s3nh/SeverusWestLake-7B-DPO. Expert labels do not independently establish total or active learned parameter counts. Provider: jsfs11. Access: Open.
Aggregate Rating 1427 ± 34, rank #4826 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 80.98 | Accuracy (%) (5-shot) | 76.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 66.38 | Normalized accuracy (%) (25-shot) | 73.4 |
| Open LLM Leaderboard v1 - HellaSwag | 84.9 | Normalized accuracy (%) (10-shot) | 70 |
| Open LLM Leaderboard v1 - GSM8K | 57.54 | Accuracy (%) (5-shot) | 69.3 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 56.83 | MC2 (%) (0-shot) | 68.9 |
| Open LLM Leaderboard v1 - MMLU | 63.38 | Accuracy (%) (5-shot) | 60.3 |
Interactive version: theaggregate.ai/model?slug=mixtureofmerges-moe-4x7b-v10-mixtralv0-3 · How It Works · Data refreshed daily, snapshot 2026-11-08.