ResearchBenchmarksMathematical Reasoning on GSM8K (∆GSM8K (pp))Follow3.5∆GSM8K (pp)Aletheia-1.8768-0.48090.9152.3109Apr 4, 2026Evaluation ResultsMethodMethodLinks∆GSM8K (pp)AletheiaModel=StableLM-2-1.6BModel=StableLM-2-1.6B2026.043.5AletheiaModel=GPT-J-6BModel=GPT-J-6B2026.042.17AletheiaModel=Llama-3.1-8BModel=Llama-3.1-8B2026.040AletheiaModel=Qwen2.5-3BModel=Qwen2.5-3B2026.04-0.67AletheiaModel=Qwen2.5-7BModel=Qwen2.5-7B2026.04-1.5AletheiaModel=Mixtral-8×7BModel=Mixtral-8×7B2026.04-1.67