CodeSOTAβOriginal site ↗
← All tasks

Language / BoolQ

BoolQ / sub-1B cohort

Passage + question → yes / no

Model size vs score

Model size versus accuracySlayerLab/Slayer149-Balanced-ARC-E-ft-45500: 149.33 million parameters; score 78.1%. SlayerLab/Slayer139-1.01b: 139.28 million parameters; score 75.0%. Qwen/Qwen2.5-0.5B-Instruct: 494.03 million parameters; score 71.9%. SlayerLab/Slayer149: 148.91 million parameters; score 68.8%. SlayerLab/Slayer-149M-Decision-v1: 149.33 million parameters; score 68.8%. SlayerLab/Slayer149-Balanced-ARC-E-ft-v2: 149.33 million parameters; score 62.5%. SlayerLab/Slayer139-1.01: 139.28 million parameters; score 56.3%. SlayerLab/Slayer139-1.02: 139.28 million parameters; score 56.3%. SlayerLab/Slayer149-ARC: 149.33 million parameters; score 50.0%. HuggingFaceTB/SmolLM2-360M-Instruct: 361.82 million parameters; score 34.4%. HuggingFaceTB/SmolLM2-135M-Instruct: 134.52 million parameters; score 21.9%. All points use the same pinned protocol.00%11420%22740%34160%45580%568100%Favourable region: model size at or below the cohort median; score at or above the cohort median.SmolLM2-135M-InstructSlayer139-1.01bSlayer149-Balanced-ARC-E-ft-45SlayerLab/Slayer149-Balanced-ARC-E-ft-45500 · 149.33M parameters · 78.1%SlayerLab/Slayer139-1.01b · 139.28M parameters · 75.0%Qwen/Qwen2.5-0.5B-Instruct · 494.03M parameters · 71.9%SlayerLab/Slayer149 · 148.91M parameters · 68.8%SlayerLab/Slayer-149M-Decision-v1 · 149.33M parameters · 68.8%SlayerLab/Slayer149-Balanced-ARC-E-ft-v2 · 149.33M parameters · 62.5%SlayerLab/Slayer139-1.01 · 139.28M parameters · 56.3%SlayerLab/Slayer139-1.02 · 139.28M parameters · 56.3%SlayerLab/Slayer149-ARC · 149.33M parameters · 50.0%HuggingFaceTB/SmolLM2-360M-Instruct · 361.82M parameters · 34.4%HuggingFaceTB/SmolLM2-135M-Instruct · 134.52M parameters · 21.9%Model parameters (millions)accuracy · higher is better
Favourable model size / score regionObserved Pareto frontier

Green uses cohort medians, not a universal quality threshold. Frontier points have no measured alternative with both smaller model size and higher score. Lines connect observed points; they do not predict intermediate models. Uncertainty is not included in frontier membership.

11 compatible measurements. Parameter counts come from the architecture instantiated from each pinned model configuration; no weights are loaded for this count. Click a point to inspect its evidence.

Measured results

Pilot comparison · first 32 valid examples · accuracy higher is better. No global SOTA claim.

SystemScoreSamplesElapsed¹Evidence
SlayerLab/Slayer149-Balanced-ARC-E-ft-45500
e7d4b8dea441
78.1%322sInspect run →
17 Oct 2026, 01:10 UTC
SlayerLab/Slayer139-1.01b
283ffb6d620b
75.0%322sInspect run →
17 Oct 2026, 01:13 UTC
Qwen/Qwen2.5-0.5B-Instruct
7ae557604adf
71.9%322sInspect run →
17 Oct 2026, 01:08 UTC
SlayerLab/Slayer149
7857dd28ec05
68.8%322sInspect run →
17 Oct 2026, 01:11 UTC
SlayerLab/Slayer-149M-Decision-v1
ccfd7d3d91f3
68.8%322sInspect run →
17 Oct 2026, 01:12 UTC
SlayerLab/Slayer149-Balanced-ARC-E-ft-v2
7d11e8fd6376
62.5%322sInspect run →
17 Oct 2026, 01:09 UTC
SlayerLab/Slayer139-1.01
2e72c696d7c5
56.3%322sInspect run →
17 Oct 2026, 01:12 UTC
SlayerLab/Slayer139-1.02
c4298d757ba3
56.3%322sInspect run →
17 Oct 2026, 01:08 UTC
SlayerLab/Slayer149-ARC
0e6793ecfb85
50.0%322sInspect run →
17 Oct 2026, 01:12 UTC
HuggingFaceTB/SmolLM2-360M-Instruct
a10cc1512eab
34.4%322sInspect run →
17 Oct 2026, 01:12 UTC
HuggingFaceTB/SmolLM2-135M-Instruct
12fd25f77366
21.9%321sInspect run →
17 Oct 2026, 01:08 UTC

¹ Evaluation elapsed time includes model loading after dataset selection. It is not serving latency or a fair speed comparison.

What this measures

Benchmark origin
Dataset / split
google/boolq / validation
Immutable dataset revision
35b264d03638db9f4ce671b711558bf7ff0f80d5
Upstream license metadata
cc-by-sa-3.0 · consult upstream terms
Protocol
codesota-sub1b-choice-v1 / reading
Selection
First 32 valid examples in the pinned split order. SNLI rows without a valid label are excluded.
Configuration
Seed 42. Greedy generation for math and code. Mean answer-token log likelihood for text classification, completion and declarative multiple-choice recipes. No model judge.
Read scoring and comparison rules →

Planned runs

1 qualified visits in 30 days · priority 1007.0. Interest changes execution order, within the approved baseline list.

SystemStatus
Qwen/Qwen2.5-0.5B-Instructcompleted
SlayerLab/Slayer139-1.02completed
HuggingFaceTB/SmolLM2-135M-Instructcompleted
SlayerLab/Slayer149-Balanced-ARC-E-ft-v2completed
SlayerLab/Slayer149-Balanced-ARC-E-ft-45500completed
SlayerLab/Slayer149completed
SlayerLab/Slayer139-1.01completed
SlayerLab/Slayer-149M-Decision-v1completed
SlayerLab/Slayer149-ARCcompleted
HuggingFaceTB/SmolLM2-360M-Instructcompleted
SlayerLab/Slayer139-1.01bcompleted