Domain leaderboards

coding

Benchmarks in this board: LiveCodeBench (80) · SWE-bench Verified (71) · SWE-bench Pro (69) · SWE-Lancer (67) · BigCodeBench (66) · Terminal-Bench (65)

1. 1Nemotron 3 UltraNVIDIA🇺🇸 USopenHerstellerangabeLiveCodeBench 89 · SWE-bench Verified 70.776
2. 2DeepSeek V4 ProDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.3 · LiveCodeBench 88.8 · HumanEval 62.873
3. 3Muse GlimmerMeta🇺🇸 USopenHerstellerangabeGPQA Diamond 83.5 · GPQA Diamond 83.5 · SWE-bench Verified 7667
4. 4GLM-5Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86 · SWE-bench Verified 77.863
5. 5DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 93.4 · HumanEval 69.5 · BigCodeBench 56.863
6. 6Nex-N2-ProNex AGI (Shanghai Innovation Inst.)🇨🇳 CNopenHerstellerangabeGPQA Diamond 82.6 · SWE-bench Verified 74.461
7. 7Gemini 3.5 FlashGoogle DeepMind🇺🇸 USSWE-bench Verified 71.859
8. 8GLM-5.2Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.256
9. 9Ling 3.0 Flash FinAnt Group (inclusionAI)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.353
10. 10GLM-5.1Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.253

reasoning

Benchmarks in this board: ARC-AGI-2 (71) · Humanity's Last Exam (67) · ZebraLogic (53) · GPQA Diamond (50) · MuSR (46) · BIG-Bench Hard (40)

Alle Werte in dieser Rangliste sind Herstellerangaben. Sie sind entsprechend abgewertet, ersetzen aber keine unabhaengige Messung.

1. 1Muse GlimmerMeta🇺🇸 USopenHerstellerangabeAIME 89.2 · AIME 89.2 · GPQA Diamond 83.584
2. 2DeepSeek V4 ProDeepSeek🇨🇳 CNopenHerstellerangabeHMMT 96.2 · GPQA Diamond 91.3 · LiveCodeBench 88.881
3. 3GLM-5Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeHMMT 96.9 · AIME 92.7 · GPQA Diamond 8679
4. 4GLM-5.2Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeAIME 99.2 · HMMT 94.4 · GPQA Diamond 91.277
5. 5GLM-5.1Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeAIME 95.3 · HMMT 94 · GPQA Diamond 86.274
6. 6DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 93.4 · HumanEval 69.5 · BigCodeBench 56.864
7. 7Nex-N2-ProNex AGI (Shanghai Innovation Inst.)🇨🇳 CNopenHerstellerangabeGPQA Diamond 82.6 · SWE-bench Verified 74.461
8. 8Ling 3.0 Flash FinAnt Group (inclusionAI)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.359
9. 9Nemotron 3 UltraNVIDIA🇺🇸 USopenHerstellerangabeLiveCodeBench 89 · SWE-bench Verified 70.759
10. 10Solar-Open2Upstage🇰🇷 KRopenHerstellerangabeAIME 80 · GPQA Diamond 66.2 · LiveCodeBench 56.555

agentic

Benchmarks in this board: GAIA (67) · BrowseComp (67) · OSWorld (63) · TAU-bench (55) · WebArena (49) · MLE-bench (45)

1. 1DeepSeek V4 ProDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.3 · LiveCodeBench 88.8 · HumanEval 62.864
2. 2Muse GlimmerMeta🇺🇸 USopenHerstellerangabeGPQA Diamond 83.5 · GPQA Diamond 83.5 · SWE-bench Verified 7660
3. 3Nemotron 3 UltraNVIDIA🇺🇸 USopenHerstellerangabeLiveCodeBench 89 · SWE-bench Verified 70.759
4. 4GLM-5Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86 · SWE-bench Verified 77.856
5. 5GLM-5.2Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.256
6. 6DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 93.4 · HumanEval 69.5 · BigCodeBench 56.854
7. 7Nex-N2-ProNex AGI (Shanghai Innovation Inst.)🇨🇳 CNopenHerstellerangabeGPQA Diamond 82.6 · SWE-bench Verified 74.454
8. 8Ling 3.0 Flash FinAnt Group (inclusionAI)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.353
9. 9GLM-5.1Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.253
10. 10Gemini 3.5 FlashGoogle DeepMind🇺🇸 USSWE-bench Verified 71.848

long-context

Benchmarks in this board: LongBench v2 (90) · RULER (68) · MRCR (53) · BABILong (53) · Needle in a Haystack (46)

Alle Werte in dieser Rangliste sind Herstellerangaben. Sie sind entsprechend abgewertet, ersetzen aber keine unabhaengige Messung.

1. 1Nemotron 3 UltraNVIDIA🇺🇸 USopenHerstellerangabeRULER 94.7 · LongBench v2 61.973
2. 2Nemotron 3.5 LightningNVIDIA🇺🇸 USopenHerstellerangabeRULER 82.3660
3. 3Muse GlimmerMeta🇺🇸 USopenHerstellerangabeGPQA Diamond 83.5 · GPQA Diamond 83.556
4. 4GLM-5.2Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.256
5. 5Ling 3.0 Flash FinAnt Group (inclusionAI)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.353
6. 6GLM-5.1Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.253
7. 7GLM-5Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 8652
8. 8Nex-N2-ProNex AGI (Shanghai Innovation Inst.)🇨🇳 CNopenHerstellerangabeGPQA Diamond 82.650
9. 9DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 93.4 · LongBench v2 44.746
10. 10DeepSeek V4 ProDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.3 · LongBench v2 40.243

knowledge

Benchmarks in this board: SimpleQA (81) · MMLU-Pro (50) · MMLU (43)

Alle Werte in dieser Rangliste sind Herstellerangaben. Sie sind entsprechend abgewertet, ersetzen aber keine unabhaengige Messung.

1. 1DeepSeek V4 ProDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.3 · MMLU 87.8 · MGSM 81.366
2. 2Nemotron 3 UltraNVIDIA🇺🇸 USopenHerstellerangabeMMLU 86.858
3. 3Motif-3Motif Technologies🇰🇷 KRopenHerstellerangabeMMLU 86.257
4. 4LongCat-Flash-Lite-SparseMeituan LongCat🇨🇳 CNopenHerstellerangabeMMLU 85.52 · GPQA Diamond 66.7857
5. 5Muse GlimmerMeta🇺🇸 USopenHerstellerangabeGPQA Diamond 83.5 · GPQA Diamond 83.556
6. 6GLM-5.2Z.ai (Zhipu)🇨🇳 CNopenHerstellerangabeGPQA Diamond 91.256
7. 7Solar-Open2Upstage🇰🇷 KRopenHerstellerangabeMMLU 80.4 · GPQA Diamond 66.255
8. 8DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeGPQA Diamond 93.4 · MGSM 84.4 · MMLU 68.354
9. 9Nemotron 3.5 LightningNVIDIA🇺🇸 USopenHerstellerangabeMMLU 81.0754
10. 10Ling 3.0 Flash FinAnt Group (inclusionAI)🇨🇳 CNopenHerstellerangabeGPQA Diamond 86.353

vision

Benchmarks in this board: MMMU-Pro (80) · DocVQA (71) · BLINK (63) · MathVista (53) · RealWorldQA (44) · AI2D (44)

Alle Werte in dieser Rangliste sind Herstellerangaben. Sie sind entsprechend abgewertet, ersetzen aber keine unabhaengige Messung.

1. 1Mage-VLMicrosoft🇺🇸 USopenHerstellerangabeDocVQA 94.69 · ChartQA 83.96 · AI2D 81.5483
2. 2Muse GlimmerMeta🇺🇸 USopenHerstellerangabeMMMU-Pro 74 · MMMU-Pro 7472
3. 3DeepSeek V4.1 FlashDeepSeek🇨🇳 CNopenHerstellerangabeDocVQA 95.6 · MMMU-Pro 56.571
4. 4Phi-4-multimodalMicrosoft🇺🇸 USopenHerstellerangabeChartQA 81.8 · AI2D 78.1 · DocVQA 69.354

audio-stt

Benchmarks in this board: Earnings-22 (68) · AMI Meeting Corpus (63) · Open ASR Leaderboard (58) · Common Voice (ASR) (52) · FLEURS (52) · TED-LIUM 3 (49)

1. 1azure-speech-07Microsoft🇺🇸 USCommon Voice (ASR) 4.32 · FLEURS 2.48 · Common Voice (ASR) 2.3698
2. 2azure-speech-06Microsoft🇺🇸 USCommon Voice (ASR) 4.28 · FLEURS 2.78 · FLEURS 2.4798
3. 3canary-v2NVIDIA🇺🇸 USopenCommon Voice (ASR) 6.58 · Common Voice (ASR) 4.69 · Common Voice (ASR) 4.5196
4. 4VoxtralMistral AI🇫🇷 FRopenCommon Voice (ASR) 9.57 · FLEURS 8.19 · Common Voice (ASR) 7.8196
5. 5muse-voice-transcribeMeta🇺🇸 USopenCommon Voice (ASR) 9.41 · FLEURS 5.94 · Common Voice (ASR) 5.395
6. 6granite-speech-4.1IBM🇺🇸 USopenCommon Voice (ASR) 6.38 · FLEURS 6.24 · FLEURS 5.2695
7. 7transcribe-03Cohere🇨🇦 CAopenEarnings-22 12.66 · Open ASR Leaderboard 9.733125 · Common Voice (ASR) 5.2895
8. 8parakeet-tdt-v3NVIDIA🇺🇸 USopenEarnings-22 14.58 · Open ASR Leaderboard 10.72 · Common Voice (ASR) 6.3595
9. 9Qwen3-ASRAlibaba Qwen🇨🇳 CNopenCommon Voice (ASR) 10.78 · Common Voice (ASR) 8.56 · Common Voice (ASR) 7.9795
10. 10omniASR-LLM-v2Meta🇺🇸 USopenCommon Voice (ASR) 12.07 · FLEURS 11.21 · Common Voice (ASR) 10.3194

Runs on 16 GB

Open-weight models that fit a 16 GB card, ranked by measured quality. The size shown is the largest variant that still fits.

1. 1VoxtralMistral AI🇫🇷 FRopen4B @ FP16 ≈ 10.96 GB96
2. 2canary-v2NVIDIA🇺🇸 USopen1B @ FP16 ≈ 3.58 GB96
3. 3granite-speech-4.1IBM🇺🇸 USopen2B @ FP16 ≈ 6.08 GB95
4. 4parakeet-tdt-v3NVIDIA🇺🇸 USopen0.6B @ FP16 ≈ 2.66 GB95
5. 5Qwen3-ASRAlibaba Qwen🇨🇳 CNopen1.7B @ FP16 ≈ 5.34 GB95
6. 6omniASR-LLM-v2Meta🇺🇸 USopen7B @ Q8 ≈ 10.12 GB94
7. 7gemma-4-E4BGoogle DeepMind🇺🇸 USopen4B @ FP16 ≈ 10.96 GB93
8. 8nemotron-3.5-asr-streamingNVIDIA🇺🇸 USopen0.6B @ FP16 ≈ 2.66 GB92
9. 9omniASR-CTC-v2Meta🇺🇸 USopen7B @ Q8 ≈ 10.12 GB91
10. 10parakeet-tdt-v2NVIDIA🇺🇸 USopen0.6B @ FP16 ≈ 2.66 GB90

Adoption

HuggingFace downloads measure spread in the open ecosystem, not capability — proprietary models cannot appear here at all. Kept as a separate question, not as the main ranking.

Data as of 2026-08-25