ARCHER Baselines: leaderboard
Released language-model baselines solve the paper’s reasoning tasks. Different training and prompting comparisons remain separate, with the source’s exact task scores and model sizes.
Source: arxiv.org.
Benchmark results
Results are published separately for each benchmark or evaluation setting.
- ARCHER Baselines - CoT-FSP - AIME 2024
- ARCHER Baselines - CoT-FSP - AIME 2025
- ARCHER Baselines - CoT-FSP - GSM8K
- ARCHER Baselines - CoT-FSP - MATH
- ARCHER Baselines - CoT-FSP - MATH500
- ARCHER Baselines - CoT-FSP - MMLU Mathematical Subset
- ARCHER Baselines - CoT-FSP - PolyMATH
- ARCHER Baselines - CoT-FSP - SVAMP
- ARCHER Baselines - Vanilla - AIME 2024
- ARCHER Baselines - Vanilla - AIME 2025
- ARCHER Baselines - Vanilla - GSM8K
- ARCHER Baselines - Vanilla - MATH
- ARCHER Baselines - Vanilla - MATH500
- ARCHER Baselines - Vanilla - MMLU Mathematical Subset
- ARCHER Baselines - Vanilla - PolyMATH
- ARCHER Baselines - Vanilla - SVAMP
Interactive version: theaggregate.ai/benchmark?slug=archer-baselines · How It Works · Data refreshed daily, snapshot 2026-11-17.