GALAXY-XB-v.03: benchmark results

An experimental depth-upscaled merge released by TeeZee from SOLAR-10.7B-v1.0, removing 12 layers from each source model before merging. This revision is a merge without subsequent fine-tuning. Provider: TeeZee. Access: Open.

Aggregate Rating 1414 ± 33, rank #5474 of 10916 rated models, from 6 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Open LLM Leaderboard v1 - WinoGrande81.06Accuracy (%) (5-shot)77
Open LLM Leaderboard v1 - MMLU64.55Accuracy (%) (5-shot)72.4
Open LLM Leaderboard v1 - HellaSwag83.59Normalized accuracy (%) (10-shot)59.9
Open LLM Leaderboard v1 - GSM8K45.03Accuracy (%) (5-shot)59.1
Open LLM Leaderboard v1 - ARC Challenge61.77Normalized accuracy (%) (25-shot)54.4
Open LLM Leaderboard v1 - TruthfulQA MC244.19MC2 (%) (0-shot)26.3

Interactive version: theaggregate.ai/model?slug=galaxy-xb-v-03 · How It Works · Data refreshed daily, snapshot 2026-11-08.