Open SLM Leaderboard - PIQA: leaderboard
The Open SLM Leaderboard (Axiomic Labs, Hugging Face Space) ranks open-weight language models under 400M parameters that their submitters pretrained from scratch, with zero-shot scores from LM-eval harness and Axiomic Labs' ArithMark-3 script that the leaderboard team checks before merging. Unlike the Hugging Face Open LLM Leaderboard, which evaluated every submitted model on its own cluster, it lists zero-shot results the submitters ran and admits only small from-scratch checkpoints (no merges). The page does not state the harness version, the accuracy variant (acc or acc_norm) or the evaluated splits. Rows are bound to their Hugging Face model card; a row without one (Mystery-5M-Preview) is withheld, and a row missing a score is absent from that board only. PIQA physical-commonsense two-choice questions, zero-shot accuracy (%, chance 50).
Metric: Zero-shot accuracy (%, chance 50). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 264 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | SmolLM2-360M (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=HuggingFaceTB/SmolLM2-360M) | 71.6 | |
| 2 | GPT-X3-150M (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=AxiomicLabs/GPT-X3-150M) | 71.33 | |
| 3 | Boris-2-Preview-2 (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=opencerebral/Boris-2-Preview-2) | 70.84 | |
| 4 | KLM-1.6 (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=MinimaLabs/KLM-1.6) | 70.18 | |
| 5 | Rose-1.5-Pro (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=GODELEV/Rose-1.5-Pro) | 69.48 | |
| 6 | GPT-X2.5-135M (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=AxiomicLabs/GPT-X2.5-135M) | 69.42 | |
| 7 | Boris-2-Preview-1 (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=opencerebral/Boris-2-Preview-1) | 69.21 | |
| 8 | BananaMind-2.5-Pro (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=BananaMind/BananaMind-2.5-Pro) | 69.02 | |
| 9 | KLM-1.5 (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=MinimaLabs/KLM-1.5) | 68.55 | |
| 10 | SmolLM2-135M (Benchmark Sweep Run: source=Open SLM Leaderboard; zero-shot; card=HuggingFaceTB/SmolLM2-135M) | 68.44 |
Interactive version: theaggregate.ai/benchmark?slug=open-slm-leaderboard-piqa · How It Works · Data refreshed daily, snapshot 2026-11-12.