🏆 Tiny League

Benchmark scores for recent small open-weight AI models (4B–190B; official weights, current versions only).

Generated 2026-09-05T19:53:15Z

Models

Click any column header to sort (Params / Benchmarks / AA-Index sort numerically, Released by date; click again to flip direction; missing values sink to the bottom). A V badge marks models that support vision / multimodal (image) input.

ModelFamilyParamsArchitectureReleasedBenchmarksAA-Index
Qwen3.8-27BVQwen3.827.8Bdense2026-08-052641
Qwen3.8-Flash-NextVQwen3.8180Bmoe2026-08-242346
Qwen-AgentWorld-35B-A3BVQwen-AgentWorld34.7Bmoe2026-06-221
gemma-4-12B-itVgemma-412Bdense2026-05-231716
diffusiongemma-26B-A4B-itVdiffusiongemma25.8Bmoe2026-06-09158
DeepSeek-V4-Flash-DSparkDeepSeek-V4165.3Bmoe2026-06-272241
Laguna-S-2.1Laguna117.6Bmoe2026-07-136
Laguna-XS-2.1Laguna33.4Bmoe2026-06-204
LFM2.5-8B-A1BLFM2.58.5Bmoe2026-05-28113
Nanbeige4.2-3BNanbeige4.24.2Bdense2026-07-2121
KAT-Coder-V2.5-DevVKAT-Coder34.7Bdense2026-07-237
North-Mini-Code-1.0North30.5Bmoe2026-06-05913.4
Mellum2-12B-A2.5B-ThinkingMellum212.1Bmoe2026-05-2612
granite-4.2-30bgranite-4.229.3Bdense2026-08-072017
granite-4.2-8bgranite-4.28.8Bdense2026-08-072014
Hy-MT2-30B-A3BHy-MT230.1Bmoe2026-05-119
LongCat-Flash-Lite-SparseLongCat69.1Bdense2026-07-313011
Fara1.5-27BVFara27.4Bdense2026-07-173
Fara1.5-9BVFara9.4Bdense2026-05-123
Nemotron-3-Nano-Omni-30B-A3B-ReasoningVNemotron-333Bmoe2026-04-20129
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16Nemotron-3.531.6Bmoe2026-08-011516
Ornith-1.5-9BVOrnith9.7Bdense2026-08-1813
Ornith-1.5-35B-A3BVOrnith36Bmoe2026-08-1815
Ling-3.0-tinyLing7.9Bmoe2026-08-101825
Ling-3.0-flashLing127.5Bmoe2026-08-022627
Spark-X2.5-4BSpark X4.1Bdense2026-08-2421

Coverage heatmap

Cell shows the score for that model×benchmark; grey = absent / no numeric score. Cell colour is scaled per benchmark (light → deep blue = lower → higher); the gold cell is the best model on that benchmark. Columns are ordered by how many models report each benchmark. Click a benchmark column to rank models by it; click the Model column to sort by name.

lowerhigherbest on benchmarkVsupports vision / multimodal
ModelGPQA DiamondSWE-bench ProTerminal-BenchAA-Intelligence-…HLESWE-bench Multil…IFBenchSWE-bench Verifi…LiveCodeBenchAIMESciCodeBFCLBrowseCompMCP-AtlasMMLU-ProAA-LCRHMMT Feb 2026IMO AnswerBenchDeepSWEIFEvalMathVisionNL2Repo-BenchToolathlonτ²-benchClawEvalCodeforces ELOGDPvalGDPval-AAMRCRMulti-IFOmniDocBench 1.5WideSearchτ³-bench
Qwen3.8-27BV89.261.7734130.8·79.5·90.3·········42.2·9042.3········91.1··
Qwen3.8-Flash-NextV91.762.5·4635.98181.3·91.9·········58.7·90.648.173.5··········
Qwen-AgentWorld-35B-A3BV·································
gemma-4-12B-itV78.8··165.2···7277.5····77.2·····79.7··69·1659····0.2··
diffusiongemma-26B-A4B-itV73.2··811···69.169.1····77.6·····70.5··56.2·1429····0.3··
DeepSeek-V4-Flash-DSpark88.152.656.94134.873.3·7991.6···73.26986.2·94.888.4····47.8··3052·139578.7····
Laguna-S-2.1·59.470.2··78.5············40.4···49.7··········
Laguna-XS-2.1·47.637.5··63.1·70.9·························
LFM2.5-8B-A1B···3··56.5··42.5·64.8·······91.8···88.1·····79.9···
Nanbeige4.2-3B87.446.944.1·17.8·54.663.672.5·35.6··57.8·58.782.867.3······52.2·74.3······
KAT-Coder-V2.5-DevV·46.041.0··63·69.4··44.2······················
North-Mini-Code-1.075.740.23613.411.1·57.567.670.3·38.2······················
Mellum2-12B-A2.5B-Thinking57.6·······69.958.4·45.6·······76.5·············
granite-4.2-30b66.433.329.217·41.977.25775.889.238.861.4··77.6···········1225·····62
granite-4.2-8b64.119.120.614·30.879.347.773.286.736.152.4··74.0···········1189·····58.1
Hy-MT2-30B-A3B······50.7············89.8·············
LongCat-Flash-Lite-Sparse69.540.633.711·59.3·68.2·65.7··48.645.679.24840.549.4··········44.7····
Fara1.5-27BV·································
Fara1.5-9BV·································
Nemotron-3-Nano-Omni-30B-A3B-ReasoningV···9·····························
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1675.4·24.61611.739.371.951.6··32.6·37.0·81.952·················
Ornith-1.5-9BV86.447.546.2·20.254.4·70.6····56.454.2·······32.4··66.5······59.5·
Ornith-1.5-35B-A3BV89.259.667.8·25.671.4·79····67.670.2····22··46.248.7·72.5······67.8·
Ling-3.0-tiny73.4·27.7259.3·63.6···24.262.7···58.770.371.0·········772·83.2···
Ling-3.0-flash·56.6572722.772.474.5·82.893.2·7372.265.5·65.18783.7·········110790.887.7·73.6·
Spark-X2.5-4B67.444.4··12.353.37541.6·90.734.765.140.954.6·56.381.274.2·93···75.1········30.4

AA-Intelligence-Index vs parameters

Comparisons

Bar charts compare models that report the same benchmark (≥2 models within the parameter range above). Hover a bar for the exact value.

Per-benchmark detail

GPQA Diamond (16 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next91.7%Hugging Face model cardlink
Qwen3.8-27B89.2%Hugging Face model cardlink
Ornith-1.5-35B-A3B89.2%Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark88.1%Pass@1Hugging Face model cardlink
Nanbeige4.2-3B87.4%Hugging Face model cardlink
Ornith-1.5-9B86.4%Hugging Face model cardlink
gemma-4-12B-it78.8%Hugging Face model cardlink
North-Mini-Code-1.075.7%OverallArtificial Analysislink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1675.4%no toolsHugging Face model cardlink
Ling-3.0-tiny73.4%Hugging Face model cardlink
diffusiongemma-26B-A4B-it73.2%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse69.5%avg@16, w/o HIHugging Face model cardlink
Spark-X2.5-4B67.4%Hugging Face model cardlink
granite-4.2-30b66.4%Hugging Face model cardlink
granite-4.2-8b64.1%Hugging Face model cardlink
Mellum2-12B-A2.5B-Thinking57.6%Hugging Face model cardlink

SWE-bench Pro (15 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next62.5%Hugging Face model cardlink
Qwen3.8-27B61.7%Hugging Face model cardlink
Ornith-1.5-35B-A3B59.6%Hugging Face model cardlink
Laguna-S-2.159.4%Public DatasetHugging Face model cardlink
Ling-3.0-flash56.6%Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark52.6%ResolvedHugging Face model cardlink
Laguna-XS-2.147.6%Public Dataset, pass@1 (avg over 2 attempts)Hugging Face model cardlink
Ornith-1.5-9B47.5%Hugging Face model cardlink
Nanbeige4.2-3B46.9%Hugging Face model cardlink
KAT-Coder-V2.5-Dev46.0%pass@1Hugging Face model cardlink
Spark-X2.5-4B44.4%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse40.6%w/o HIHugging Face model cardlink
North-Mini-Code-1.040.2%OverallHugging Face model cardlink
granite-4.2-30b33.3%Hugging Face model cardlink
granite-4.2-8b19.1%Hugging Face model cardlink

Terminal-Bench (15 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-27B73%2.1 (Terminus)Hugging Face model cardlink
Laguna-S-2.170.2%2.1Hugging Face model cardlink
Ornith-1.5-35B-A3B67.8%2.1 (Terminus-2)Hugging Face model cardlink
Ling-3.0-flash57%2.1Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark56.9%2.0Hugging Face model cardlink
Ornith-1.5-9B46.2%2.1 (Terminus-2)Hugging Face model cardlink
Nanbeige4.2-3B44.1%2.0Hugging Face model cardlink
KAT-Coder-V2.5-Dev41.0%v2.1, avg of two harnessesHugging Face model cardlink
Laguna-XS-2.137.5%v2.0, pass@1 (avg over 5 attempts)Hugging Face model cardlink
North-Mini-Code-1.036%v2Hugging Face model cardlink
LongCat-Flash-Lite-Sparse33.7%2.0, w/o HIHugging Face model cardlink
granite-4.2-30b29.2%2.1Hugging Face model cardlink
Ling-3.0-tiny27.7%2.1Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1624.6%v2.1Hugging Face model cardlink
granite-4.2-8b20.6%2.1Hugging Face model cardlink

AA-Intelligence-Index (14 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next46idxArtificial Analysislink
Qwen3.8-27B41indexreasoning level: xhigh (highest)Artificial Analysislink
DeepSeek-V4-Flash-DSpark41idxArtificial Analysislink
Ling-3.0-flash27idxArtificial Analysislink
Ling-3.0-tiny25idxv4.1.1Hugging Face model cardlink
granite-4.2-30b17indexArtificial Analysislink
gemma-4-12B-it16indexArtificial Analysislink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1616indexArtificial Analysislink
granite-4.2-8b14indexArtificial Analysislink
North-Mini-Code-1.013.4idxv4.2Artificial Analysislink
LongCat-Flash-Lite-Sparse11indexArtificial Analysislink
Nemotron-3-Nano-Omni-30B-A3B-Reasoning9indexArtificial Analysislink
diffusiongemma-26B-A4B-it8indexArtificial Analysislink
LFM2.5-8B-A1B3idxArtificial Analysislink

HLE (13 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next35.9%Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark34.8%Pass@1Hugging Face model cardlink
Qwen3.8-27B30.8%Hugging Face model cardlink
Ornith-1.5-35B-A3B25.6%no toolsHugging Face model cardlink
Ling-3.0-flash22.7%Hugging Face model cardlink
Ornith-1.5-9B20.2%no toolsHugging Face model cardlink
Nanbeige4.2-3B17.8%w/o searchHugging Face model cardlink
Spark-X2.5-4B12.3%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1611.7%text-only, no toolsHugging Face model cardlink
North-Mini-Code-1.011.1%OverallArtificial Analysislink
diffusiongemma-26B-A4B-it11%no toolsHugging Face model cardlink
Ling-3.0-tiny9.3%Hugging Face model cardlink
gemma-4-12B-it5.2%no toolsHugging Face model cardlink

SWE-bench Multilingual (13 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next81%Hugging Face model cardlink
Laguna-S-2.178.5%Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark73.3%ResolvedHugging Face model cardlink
Ling-3.0-flash72.4%Hugging Face model cardlink
Ornith-1.5-35B-A3B71.4%Hugging Face model cardlink
Laguna-XS-2.163.1%pass@1 (avg over 4 attempts)Hugging Face model cardlink
KAT-Coder-V2.5-Dev63%pass@1Hugging Face model cardlink
LongCat-Flash-Lite-Sparse59.3%w/o HIHugging Face model cardlink
Ornith-1.5-9B54.4%Hugging Face model cardlink
Spark-X2.5-4B53.3%Hugging Face model cardlink
granite-4.2-30b41.9%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1639.3%OverallHugging Face model cardlink
granite-4.2-8b30.8%Hugging Face model cardlink

IFBench (12 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next81.3%Hugging Face model cardlink
Qwen3.8-27B79.5%Hugging Face model cardlink
granite-4.2-8b79.3%promptHugging Face model cardlink
granite-4.2-30b77.2%promptHugging Face model cardlink
Spark-X2.5-4B75%Hugging Face model cardlink
Ling-3.0-flash74.5%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1671.9%looseHugging Face model cardlink
Ling-3.0-tiny63.6%Hugging Face model cardlink
North-Mini-Code-1.057.5%OverallArtificial Analysislink
LFM2.5-8B-A1B56.5%Hugging Face model cardlink
Nanbeige4.2-3B54.6%Hugging Face model cardlink
Hy-MT2-30B-A3B50.7%OverallOfficial tech report / arXiv paperlink

SWE-bench Verified (12 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark79%ResolvedHugging Face model cardlink
Ornith-1.5-35B-A3B79%Hugging Face model cardlink
Laguna-XS-2.170.9%pass@1 (avg over 4 attempts)Hugging Face model cardlink
Ornith-1.5-9B70.6%Hugging Face model cardlink
KAT-Coder-V2.5-Dev69.4%pass@1Hugging Face model cardlink
LongCat-Flash-Lite-Sparse68.2%w/o HIHugging Face model cardlink
North-Mini-Code-1.067.6%OverallHugging Face model cardlink
Nanbeige4.2-3B63.6%Hugging Face model cardlink
granite-4.2-30b57%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1651.6%OverallHugging Face model cardlink
granite-4.2-8b47.7%Hugging Face model cardlink
Spark-X2.5-4B41.6%Hugging Face model cardlink

LiveCodeBench (11 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next91.9%v6Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark91.6%Pass@1Hugging Face model cardlink
Qwen3.8-27B90.3%v6Hugging Face model cardlink
Ling-3.0-flash82.8%2408-2505Hugging Face model cardlink
granite-4.2-30b75.8%v6Hugging Face model cardlink
granite-4.2-8b73.2%v6Hugging Face model cardlink
Nanbeige4.2-3B72.5%v6Hugging Face model cardlink
gemma-4-12B-it72%v6Hugging Face model cardlink
North-Mini-Code-1.070.3%v6Hugging Face model cardlink
Mellum2-12B-A2.5B-Thinking69.9%v6Hugging Face model cardlink
diffusiongemma-26B-A4B-it69.1%v6Hugging Face model cardlink

AIME (9 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash93.2%2026Hugging Face model cardlink
Spark-X2.5-4B90.7%2026Hugging Face model cardlink
granite-4.2-30b89.2%2025Hugging Face model cardlink
granite-4.2-8b86.7%2025Hugging Face model cardlink
gemma-4-12B-it77.5%2026 no toolsHugging Face model cardlink
diffusiongemma-26B-A4B-it69.1%2026 no toolsHugging Face model cardlink
LongCat-Flash-Lite-Sparse65.7%2026, avg@32, w/o HIHugging Face model cardlink
Mellum2-12B-A2.5B-Thinking58.4%Hugging Face model cardlink
LFM2.5-8B-A1B42.5%2025Hugging Face model cardlink

SciCode (8 models)

ModelScoreMetricSubtaskSourceURL
KAT-Coder-V2.5-Dev44.2%pass@1Hugging Face model cardlink
granite-4.2-30b38.8%Hugging Face model cardlink
North-Mini-Code-1.038.2%OverallHugging Face model cardlink
granite-4.2-8b36.1%Hugging Face model cardlink
Nanbeige4.2-3B35.6%Hugging Face model cardlink
Spark-X2.5-4B34.7%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1632.6%OverallHugging Face model cardlink
Ling-3.0-tiny24.2%Hugging Face model cardlink

BFCL (7 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash73%v4Hugging Face model cardlink
Spark-X2.5-4B65.1%v4Hugging Face model cardlink
LFM2.5-8B-A1B64.8%v3Hugging Face model cardlink
Ling-3.0-tiny62.7%v4Hugging Face model cardlink
granite-4.2-30b61.4%v4Hugging Face model cardlink
granite-4.2-8b52.4%v4Hugging Face model cardlink
Mellum2-12B-A2.5B-Thinking45.6%v4Hugging Face model cardlink

BrowseComp (7 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark73.2%Pass@1Hugging Face model cardlink
Ling-3.0-flash72.2%single-agent (w/ ctx)Hugging Face model cardlink
Ornith-1.5-35B-A3B67.6%Hugging Face model cardlink
Ornith-1.5-9B56.4%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse48.6%pass@1, w/o HIHugging Face model cardlink
Spark-X2.5-4B40.9%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1637.0%OverallHugging Face model cardlink

MCP-Atlas (7 models)

ModelScoreMetricSubtaskSourceURL
Ornith-1.5-35B-A3B70.2%Hugging Face model cardlink
DeepSeek-V4-Flash-DSpark69%Pass@1Hugging Face model cardlink
Ling-3.0-flash65.5%Hugging Face model cardlink
Nanbeige4.2-3B57.8%Hugging Face model cardlink
Spark-X2.5-4B54.6%Hugging Face model cardlink
Ornith-1.5-9B54.2%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse45.6%w/o HIHugging Face model cardlink

MMLU-Pro (7 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark86.2%EMHugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1681.9%OverallHugging Face model cardlink
LongCat-Flash-Lite-Sparse79.2%w/o HIHugging Face model cardlink
diffusiongemma-26B-A4B-it77.6%Hugging Face model cardlink
granite-4.2-30b77.6%Hugging Face model cardlink
gemma-4-12B-it77.2%Hugging Face model cardlink
granite-4.2-8b74.0%Hugging Face model cardlink

AA-LCR (6 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash65.1%Hugging Face model cardlink
Nanbeige4.2-3B58.7%Hugging Face model cardlink
Ling-3.0-tiny58.7%Hugging Face model cardlink
Spark-X2.5-4B56.3%Hugging Face model cardlink
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF1652idxOverallHugging Face model cardlink
LongCat-Flash-Lite-Sparse48%w/o HIHugging Face model cardlink

HMMT Feb 2026 (6 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark94.8%Pass@1Hugging Face model cardlink
Ling-3.0-flash87%Hugging Face model cardlink
Nanbeige4.2-3B82.8%Hugging Face model cardlink
Spark-X2.5-4B81.2%Hugging Face model cardlink
Ling-3.0-tiny70.3%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse40.5%avg@32, w/o HIHugging Face model cardlink

IMO AnswerBench (6 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark88.4%Pass@1Hugging Face model cardlink
Ling-3.0-flash83.7%Hugging Face model cardlink
Spark-X2.5-4B74.2%Hugging Face model cardlink
Ling-3.0-tiny71.0%Hugging Face model cardlink
Nanbeige4.2-3B67.3%Hugging Face model cardlink
LongCat-Flash-Lite-Sparse49.4%avg@4, w/o HIHugging Face model cardlink

DeepSWE (4 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next58.7%1.1Hugging Face model cardlink
Qwen3.8-27B42.2%1.1Hugging Face model cardlink
Laguna-S-2.140.4%Hugging Face model cardlink
Ornith-1.5-35B-A3B22%Hugging Face model cardlink

IFEval (4 models)

ModelScoreMetricSubtaskSourceURL
Spark-X2.5-4B93%Hugging Face model cardlink
LFM2.5-8B-A1B91.8%Hugging Face model cardlink
Hy-MT2-30B-A3B89.8%OverallOfficial tech report / arXiv paperlink
Mellum2-12B-A2.5B-Thinking76.5%Hugging Face model cardlink

MathVision (4 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next90.6%without CIHugging Face model cardlink
Qwen3.8-27B90%without CIHugging Face model cardlink
gemma-4-12B-it79.7%Hugging Face model cardlink
diffusiongemma-26B-A4B-it70.5%Hugging Face model cardlink

NL2Repo-Bench (4 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next48.1%Hugging Face model cardlink
Ornith-1.5-35B-A3B46.2%Hugging Face model cardlink
Qwen3.8-27B42.3%Hugging Face model cardlink
Ornith-1.5-9B32.4%Hugging Face model cardlink

Toolathlon (4 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-Flash-Next73.5%Verified, Pass@1Hugging Face model cardlink
Laguna-S-2.149.7%VerifiedHugging Face model cardlink
Ornith-1.5-35B-A3B48.7%VerifiedHugging Face model cardlink
DeepSeek-V4-Flash-DSpark47.8%Pass@1Hugging Face model cardlink

τ²-bench (4 models)

ModelScoreMetricSubtaskSourceURL
LFM2.5-8B-A1B88.1%TelecomHugging Face model cardlink
Spark-X2.5-4B75.1%Hugging Face model cardlink
gemma-4-12B-it69%average over 3Hugging Face model cardlink
diffusiongemma-26B-A4B-it56.2%average over 3Hugging Face model cardlink

ClawEval (3 models)

ModelScoreMetricSubtaskSourceURL
Ornith-1.5-35B-A3B72.5%Hugging Face model cardlink
Ornith-1.5-9B66.5%Hugging Face model cardlink
Nanbeige4.2-3B52.2%pass^3Hugging Face model cardlink

Codeforces ELO (3 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark3052ratingHugging Face model cardlink
gemma-4-12B-it1659ratingHugging Face model cardlink
diffusiongemma-26B-A4B-it1429ratingHugging Face model cardlink

GDPval (3 models)

ModelScoreMetricSubtaskSourceURL
granite-4.2-30b1225scoreHugging Face model cardlink
granite-4.2-8b1189pointsHugging Face model cardlink
Nanbeige4.2-3B74.3%Hugging Face model cardlink

GDPval-AA (3 models)

ModelScoreMetricSubtaskSourceURL
DeepSeek-V4-Flash-DSpark1395EloHugging Face model cardlink
Ling-3.0-flash1107pointsv2Hugging Face model cardlink
Ling-3.0-tiny772pointsv2Hugging Face model cardlink

MRCR (3 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash90.8%128KHugging Face model cardlink
DeepSeek-V4-Flash-DSpark78.7%1MHugging Face model cardlink
LongCat-Flash-Lite-Sparse44.7%8-needle, w/o HIHugging Face model cardlink

Multi-IF (3 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash87.7%Hugging Face model cardlink
Ling-3.0-tiny83.2%Hugging Face model cardlink
LFM2.5-8B-A1B79.9%Hugging Face model cardlink

OmniDocBench 1.5 (3 models)

ModelScoreMetricSubtaskSourceURL
Qwen3.8-27B91.1%Hugging Face model cardlink
diffusiongemma-26B-A4B-it0.3avg edit distanceHugging Face model cardlink
gemma-4-12B-it0.2avg edit distanceHugging Face model cardlink

WideSearch (3 models)

ModelScoreMetricSubtaskSourceURL
Ling-3.0-flash73.6%Hugging Face model cardlink
Ornith-1.5-35B-A3B67.8%Hugging Face model cardlink
Ornith-1.5-9B59.5%Hugging Face model cardlink

τ³-bench (3 models)

ModelScoreMetricSubtaskSourceURL
granite-4.2-30b62%avgHugging Face model cardlink
granite-4.2-8b58.1%avgHugging Face model cardlink
Spark-X2.5-4B30.4%Hugging Face model cardlink