Lara Mini Base Report (base-model protocol) - MGSM (8 (CoT)-shot): leaderboard

Harness: lm-evaluation-harness 0.4.14.dev0; task configs unchanged except shots (A4.T17); no chat template; identical prompts/settings/answer extraction for every model (§5.1; Appendix D). Backends: Log-likelihood tasks: HF Transformers backend, fp32 log-softmax. Generative tasks: vLLM, bf16, greedy decoding, 4,096-token context; backends agree within 0.6 points on checked tasks (§5.1; Appendix D). Shots and metrics: Per A4.T17 (e.g. HellaSwag 10-shot acc_norm, MMLU 5-shot, MMLU-Pro 5-shot CoT exact match, GSM8K 5-shot EM, MATH 4-shot Math-Verify, MGSM 8-shot CoT over 11 languages, MBPP/MBPP+ 3-shot pass@1, CRUXEval 2-shot pass@1) (A4.T17). Answer extraction: Stricter parsers for GSM8K, MGSM (native CoT answer phrase) and MBPP+ (must pass MBPP and EvalPlus tests); MATH Minerva prompts with Math-Verify over 5,000 problems; CRUXEval direct prediction with official parser (Appendix D Answer extraction). Evaluated Lara checkpoint: Released checkpoint after the 100B-token long-context stage (translated/Lara-4-Mini-Base @ c998c0e1251a734eeff364d532330790a4f93c65 as captured; revision used by the authors not stated) (§5 introduction; §7 Release). Long context: RULER (13 tasks), ONERULER (4 retrieval tasks, 26 languages), HELMET (4 categories) at 32k and 128k with base-model modifications in A4.T24 (§5.7; Appendix D.3; A4.T24). Runs / seeds: not stated; scoring is log-likelihood or greedy generation (not stated). Benchmark availability: T-HellaSwag is the authors' translation, announced as "will release" at v1 (§5.2; §7). This board: Full benchmark (MATH: 5,000 problems; others not printed); Single greedy generation per item (vLLM bf16, 4,096-token context); run count not stated. lm-evaluation-harness 0.4.14.dev0, 8 (CoT)-shot (A4.T17); answer-extraction overrides in Appendix D. Metric: Exact match on each language's native CoT answer phrase, averaged over all 11 languages (%) Higher is better.

Metric: Exact match on each language's native CoT answer phrase, averaged over all 11 languages (%). Source: arxiv.org. Saturation forecast: Around February 2027. 55 models tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1Qwen3-8B-Base71.9#1774
2Qwen3-4B-Base66.9#2469
3Hunyuan-4B-Pretrain63.1#2455
4Qwen3.5-4B-Base61.6#2478
5Qwen 2.5 7B60.8#2829
6Apertus-v1.5-8B-base54.6#4427
7Gemma 4 E4B52.5#2412
8Lara-4-Mini-Base51.7#5152
9Qwen3-1.7B-Base50.9#5246
10Ministral-3-3B-Base-251249.2#3138
11granite-4.1-3B-base47.3#6109
12Domyn-Edge-Base47#4368
13LFM3-1.7B-Base46.2#6594
14Nemotron-Flash-3B46.1#6435
15Falcon-H1-1.5B-Deep-Base44.3#5677

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=lara-mini-base-report-base-model-protocol-mgsm-8-cot-shot · How It Works · Data refreshed daily, snapshot 2026-11-16.