Use cases › Capability
Reason through hard problems
Models that think before answering and are built for multi-step reasoning, mathematics and graduate-level science questions.
Benchmarks that represent this job 17
A model's row on one of these third-party boards is what earns the measured label; the same benchmark name in the model's own card earns author-reported.
AIME 2025AIME 2026AIME 2024HMMTGPQA DiamondHumanity's Last ExamMMLU-ProMMLUMATH-500GSM8KARC-AGILiveBenchSimpleQABBHSuperGPQABBEHIMO-AnswerBench
Models 293
Narrow by what you can run and pay for. Rows are ordered measured → author-reported → structural, then by watch-list score; click a column to re-sort. Every chip links to its source.
| Model | Evidence | Params | Context | Price from | Runs on | License | Providers | Measured latency |
|---|---|---|---|---|---|---|---|---|
| GLM-5.3-Flash zai-org | author-reported Author reports hle 55.3decision-model estimate p=0.86 +1 | 321B | 1M | $0.02 | Cluster | MIT | 9 | 8.1 s · 153 tok/s |
| gemma-4-26B-A4B-it google | author-reported Author reports MMLU Pro 82.6%Author reports AIME 2026 no tools 88.3% +1 | 25.8B (4B active) | 256K | $0.00 | Single high-end GPU | Apache 2.0 | 6 | 0.9 s · 51 tok/s |
| GLM-5.3 zai-org | author-reported Author reports hle 62.5chat template: <think> | 753B | 1M | $0.07 | Cluster | GLM License (glm-5.3) review | 9 | — |
| Kimi-K3 moonshotai | author-reported Author reports GPQA Diamond 93.5Author reports hle 56 +1 | 2.78T | 1M | $0.23 | Cluster | Kimi K3 (Modified MIT) (k… gated-threshold | 7 | 25.6 s · 59 tok/s |
| gemma-4-31B-it google | author-reported Author reports MMLU Pro 85.2%Author reports AIME 2026 no tools 89.2% +1 | 31.3B | 256K | $0.00 | Single high-end GPU | Apache 2.0 | 5 | — |
| GLM-5.2 zai-org | author-reported Author reports HLE 40.5Author reports AIME 2026 99.2 +1 | 753B | 1M | $0.03 | Cluster | MIT | 9 | — |
| DeepSeek-V4.1-Flash deepseek-ai | author-reported Author reports MMLU-Pro (EM) 74.1Author reports SimpleQA-Verified (EM) 42.3 +1 | 763B | 1M | $0.02 | Cluster | MIT | 8 | — |
| Qwen3.5-9B Qwen | author-reported Author reports MMLU-Pro 82.5Author reports MMLU-Redux 91.1 +1 | 9.65B | 256K | $0.08 | Single GPU | Apache 2.0 | 6 | 2.8 s · 117 tok/s |
| DeepSeek-V4-Pro deepseek-ai | author-reported Author reports MMLU-Redux (EM) 90.8Author reports MMLU-Pro (EM) 87.5 +1 | 1.60T | 1M | $0.21 | Cluster | MIT | 5 | — |
| DeepSeek-R1 deepseek-ai | author-reported Author reports MMLU-Redux (EM) 92.9Author reports MMLU-Pro (EM) 84.0 +1 | 684B | 160K | $0.70 | Cluster | MIT | 3 | — |
| GLM-4.7-Flash zai-org | author-reported Author reports AIME 25 91.6Author reports GPQA 75.2 +1 | 31.2B | 198K | $0.06 | Single high-end GPU | MIT | 5 | — |
| Qwen3.5-397B-A17B Qwen | author-reported Author reports MMLU-Pro 87.8Author reports MMLU-Redux 94.9 +1 | 403B (17B active) | 256K | $0.39 | Cluster | Apache 2.0 | 7 | — |
| Kimi-K2.6 moonshotai | author-reported Author reports MathArena/aime_2026 96.4Author reports hle 54.0 +1 | 1.03T | 256K | $0.45 | Cluster | Kimi (Modified MIT) (modi… gated-threshold | 5 | — |
| DeepSeek-V4-Flash deepseek-ai | author-reported Author reports MMLU-Redux (EM) 89.4Author reports MMLU-Pro (EM) 68.3 +1 | 291B | 1M | $0.03 | Cluster | MIT | 4 | — |
| gemma-3-4b-it google | author-reported Author reports BIG-Bench Hard 50.9Author reports MMLU 59.6 +1 | 4.30B | 128K | $0.04 | Single GPU | Gemma Terms of Use attribution | 4 | — |
| Kimi-K2.5 moonshotai | author-reported Author reports AIME 2025 96.1Author reports HMMT 2025 (Feb) 95.4 +1 | 1.03T | 256K | $0.45 | Cluster | Kimi (Modified MIT) (modi… gated-threshold | 4 | — |
| GLM-5 zai-org | author-reported Author reports HLE 30.5Author reports AIME 2026 I 92.7 +1 | 754B | 198K | $0.60 | Cluster | MIT | 5 | — |
| Qwen3.5-35B-A3B Qwen | author-reported Author reports MMLU-Pro 85.3Author reports MMLU-Redux 93.3 +1 | 36.0B (3B active) | 256K | $0.07 | Multi-GPU node | Apache 2.0 | 4 | — |
| gemma-3-27b-it google | author-reported Author reports BIG-Bench Hard 77.7Author reports MMLU 78.6 +1 | 27.4B | 128K | $0.08 | Single high-end GPU | Gemma Terms of Use attribution | 4 | — |
| Qwen3.5-27B Qwen | author-reported Author reports MMLU-Pro 86.1Author reports MMLU-Redux 93.2 +1 | 27.8B | 256K | $0.20 | Single high-end GPU | Apache 2.0 | 4 | — |
| DeepSeek-R1-0528 deepseek-ai | author-reported Author reports MMLU-Redux (EM) 93.4Author reports MMLU-Pro (EM) 85.0 +1 | 685B | 160K | $0.50 | Cluster | MIT | 4 | — |
| MiniMax-M2.5 MiniMaxAI | author-reported Author reports hle 19.4chat template: <think> | 229B | 192K | $0.27 | Multi-GPU node | MiniMax (Modified MIT) (m… review | 4 | — |
| Qwen3.5-4B Qwen | author-reported Author reports MMLU-Pro 79.1Author reports MMLU-Redux 88.8 +1 | 4.66B | 256K | — | Single GPU | Apache 2.0 | 2 | — |
| DeepSeek-V3 deepseek-ai | author-reported Author reports MMLU-Redux (EM) 89.1Author reports MMLU-Pro (EM) 75.9 +1 | 684B | 160K | $0.26 | Cluster | Not declared review | 2 | — |
| Qwen3-Next-80B-A3B-Instruct Qwen | author-reported Author reports MMLU-Pro 80.6Author reports MMLU-Redux 90.9 +1 | 81.3B (3B active) | 256K | $0.07 | Multi-GPU node | Apache 2.0 | 4 | — |
| gemma-4-E4B-it google | author-reported Author reports MMLU Pro 69.4%Author reports AIME 2026 no tools 42.5% +1 | 8.00B | 128K | — | Single GPU | Apache 2.0 | 1 | — |
| GLM-5.1 zai-org | author-reported Author reports HLE 31.0Author reports AIME 2026 95.3 +1 | 754B | 198K | $0.96 | Cluster | MIT | 4 | — |
| phi-4 microsoft | author-reported Author reports MMLU 84.8Author reports GPQA 56.1 +1 | 14.7B | 16K | $0.06 | Single high-end GPU | MIT | 3 | — |
| gemma-3-12b-it google | author-reported Author reports BIG-Bench Hard 72.6Author reports MMLU 74.5 +1 | 12.2B | 128K | $0.04 | Single high-end GPU | Gemma Terms of Use attribution | 3 | — |
| Qwen3-4B-Instruct-2507 Qwen | author-reported Author reports MMLU-Pro 69.6Author reports MMLU-Redux 84.2 +1 | 4.02B | 256K | $0.01 | Single GPU | Apache 2.0 | 1 | — |
| gemma-3-1b-it google | author-reported Author reports BIG-Bench Hard 28.4 | 1000M | 32K | — | Laptop / edge | Gemma Terms of Use attribution | 1 | — |
| Qwen3-235B-A22B-Instruct-2507 Qwen | author-reported Author reports MMLU-Pro 83.0Author reports MMLU-Redux 93.1 +1 | 235B (22B active) | 256K | $0.09 | Multi-GPU node | Apache 2.0 | 5 | — |
| GLM-4.7 zai-org | author-reported Author reports MMLU-Pro 84.3Author reports GPQA-Diamond 85.7 +1 | 358B | 198K | $0.32 | Cluster | MIT | 4 | — |
| Qwen3-30B-A3B-Instruct-2507 Qwen | author-reported Author reports MMLU-Pro 78.4Author reports MMLU-Redux 89.3 +1 | 30.5B (3B active) | 256K | $0.09 | Single high-end GPU | Apache 2.0 | 2 | — |
| DeepSeek-R1-Distill-Qwen-32B deepseek-ai | author-reported Author reports AIME 2024 cons@64 83.3Author reports MATH-500 pass@1 94.3 +1 | 32.8B | 128K | — | Single high-end GPU | MIT | 2 | — |
| DeepSeek-R1-Distill-Qwen-1.5B deepseek-ai | author-reported Author reports AIME 2024 cons@64 52.7Author reports MATH-500 pass@1 83.9 +1 | 1.78B | 128K | — | Single GPU | MIT | 1 | — |
| DeepSeek-R1-Distill-Qwen-7B deepseek-ai | author-reported Author reports AIME 2024 cons@64 83.3Author reports MATH-500 pass@1 92.8 +1 | 7.62B | 128K | $0.15 | Single GPU | MIT | 2 | — |
| Qwen3.5-122B-A10B Qwen | author-reported Author reports MMLU-Pro 86.7Author reports MMLU-Redux 94.0 +1 | 125B (10B active) | 256K | $0.26 | Multi-GPU node | Apache 2.0 | 3 | — |
| Muse-Glimmer-30B meta-models | author-reported Author reports AIME 2026 94.7Author reports GPQA Diamond (AA) 83.5 +1 | 29.8B | 128K | $0.30 | Single high-end GPU | Apache 2.0 | 4 | — |
| gemma-4-E2B-it google | author-reported Author reports MMLU Pro 60.0%Author reports AIME 2026 no tools 37.5% +1 | 5.12B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| Qwen2-7B-Instruct Qwen | author-reported Author reports MMLU 70.5Author reports MMLU-Pro 44.1 +1 | 7.62B | 32K | — | Single GPU | Apache 2.0 | 2 | — |
| Qwen3.5-2B Qwen | author-reported Author reports MMLU-Pro 66.5Author reports MMLU-Redux 79.6 +1 | 2.27B | 256K | — | Single GPU | Apache 2.0 | 2 | — |
| Qwen3-4B-Thinking-2507 Qwen | author-reported Author reports MMLU-Pro 74.0Author reports MMLU-Redux 86.1 +1 | 4.02B | 256K | $0.01 | Single GPU | Apache 2.0 | 2 | — |
| Qwen3.5-0.8B Qwen | author-reported Author reports MMLU-Pro 42.3Author reports MMLU-Redux 59.5 +1 | 873M | 256K | — | Laptop / edge | Apache 2.0 | 1 | — |
| gemma-4-12B-it google | author-reported Author reports MMLU Pro 77.2%Author reports AIME 2026 no tools 77.5% +1 | 12.0B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| Qwen2.5-VL-32B-Instruct Qwen | author-reported Author reports MMLU 78.4Author reports MMLU-PRO 68.8 +1 | 33.5B | 125K | — | Multi-GPU node | Apache 2.0 | 2 | — |
| DeepSeek-R1-Distill-Qwen-14B deepseek-ai | author-reported Author reports AIME 2024 cons@64 80.0Author reports MATH-500 pass@1 93.9 +1 | 14.8B | 128K | $0.20 | Single high-end GPU | MIT | 2 | — |
| Step-3.5-Flash stepfun-ai | author-reported Author reports AIME 2025 97.3Author reports HMMT 2025 (Feb.) 98.4 +1 | 199B | 256K | $0.10 | Multi-GPU node | Apache 2.0 | 3 | — |
| DeepSeek-R1-Distill-Llama-8B deepseek-ai | author-reported Author reports AIME 2024 cons@64 80.0Author reports MATH-500 pass@1 89.1 +1 | 8.03B | 128K | $0.05 | Single GPU | MIT | 2 | — |
| Inkling thinkingmachines | author-reported Author reports HLE (text only) 29.7%Author reports AIME 2026 97.1% +1 | 952B | 1M | $0.00 | Cluster | Apache 2.0 | 4 | — |
| gemma-2-2b google | author-reported Author reports MMLU 51.3 | 2.61B | 8K | — | Single GPU | Gemma Terms of Use attribution | 1 | — |
| DeepSeek-R1-Distill-Llama-70B deepseek-ai | author-reported Author reports AIME 2024 pass@1 70.0decision-model estimate p=1.00 +1 | 70.6B | 128K | — | Multi-GPU node | MIT | 2 | — |
| Kimi-K2-Thinking moonshotai | author-reported Author reports HLE (Text-only) 51.0Author reports AIME25 100.0 +1 | 1.03T | 256K | $0.60 | Cluster | Kimi (Modified MIT) (modi… gated-threshold | 3 | — |
| DeepSeek-R1-0528-Qwen3-8B deepseek-ai | author-reported Author reports AIME 24 86.0Author reports AIME 25 76.3 +1 | 8.19B | 128K | — | Single GPU | MIT | 1 | — |
| Phi-3-mini-4k-instruct microsoft | author-reported Author reports GPQA 30.6Author reports MMLU 70.9 | 3.82B | 4K | — | Single GPU | MIT | 1 | — |
| diffusiongemma-26B-A4B-it google | author-reported Author reports MMLU Pro 77.6%Author reports AIME 2026 no tools 69.1% +1 | 25.8B (4B active) | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| gemma-4-31B google | author-reported Author reports MMLU Pro 85.2%Author reports AIME 2026 no tools 89.2% +1 | 31.3B | 256K | $0.00 | Single high-end GPU | Apache 2.0 | 1 | — |
| gemma-3-270m google | author-reported Author reports BIG-Bench Hard 26.7 | 268M | 32K | — | Laptop / edge | Gemma Terms of Use attribution | 0 | — |
| Phi-4-mini-instruct microsoft | author-reported Author reports BigBench Hard (0-shot, CoT) 70.4Author reports MMLU (5-shot) 67.3 +1 | 3.84B | 128K | — | Single GPU | MIT | 0 | — |
| gemma-4-E4B google | author-reported Author reports MMLU Pro 69.4%Author reports AIME 2026 no tools 42.5% +1 | 8.00B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| SmolLM2-135M HuggingFaceTB | author-reported Author reports MMLU (cloze) 31.5Author reports GSM8K (5-shot) 1.4 | 135M | 8K | — | Laptop / edge | Apache 2.0 | 0 | — |
| gemma-4-26B-A4B google | author-reported Author reports MMLU Pro 82.6%Author reports AIME 2026 no tools 88.3% +1 | 25.8B (4B active) | 256K | $0.00 | Single high-end GPU | Apache 2.0 | 2 | — |
| Hermes-3-Llama-3.1-8B NousResearch | author-reported Author reports MATH Lvl 5 (4-Shot) 4.76Author reports GPQA (0-shot) 6.38 +1 | 8.03B | 128K | — | Single GPU | Llama 3 Community License attribution | 1 | — |
| SmolLM3-3B HuggingFaceTB | author-reported Author reports AIME 2025 36.7Author reports GPQA Diamond 41.7 +1 | 3.08B | 64K | — | Single GPU | Apache 2.0 | 0 | — |
| Phi-3-mini-128k-instruct microsoft | author-reported Author reports GPQA 29.7Author reports MMLU 69.7 | 3.82B | 128K | — | Single GPU | MIT | 0 | — |
| Qwen2-72B-Instruct Qwen | author-reported Author reports MMLU 82.3Author reports MMLU-Pro 64.4 +1 | 72.7B | 32K | — | Multi-GPU node | tongyi-qianwen review | 2 | — |
| LightOnOCR-2-1B lightonai | author-reported Author reports arxiv_math 89.6 | 1.01B | 16K | — | Laptop / edge | Apache 2.0 | 1 | — |
| DeepSeek-V3.2-Exp deepseek-ai | author-reported Author reports MMLU-Pro 85.0Author reports GPQA-Diamond 79.9 +1 | 685B | 160K | $0.27 | Cluster | MIT | 2 | — |
| Qwen2-1.5B-Instruct Qwen | author-reported Author reports MMLU 52.4Author reports GSM8K 61.6 | 1.54B | 32K | — | Laptop / edge | Apache 2.0 | 1 | — |
| NVIDIA-Nemotron-Nano-9B-v2 nvidia | author-reported Author reports AIME25 72.1%Author reports MATH500 97.8% +1 | 8.89B | 128K | — | Single GPU | NVIDIA Open Model License… attribution | 1 | — |
| Qwen3-235B-A22B-Thinking-2507 Qwen | author-reported Author reports MMLU-Pro 84.4Author reports MMLU-Redux 93.8 +1 | 235B (22B active) | 256K | $0.23 | Multi-GPU node | Apache 2.0 | 3 | — |
| gemma-2-9b google | author-reported Author reports MMLU 71.3Author reports GSM8K 68.6 | 9.24B | 8K | — | Single GPU | Gemma Terms of Use attribution | 1 | — |
| gemma-3-270m-it google | author-reported Author reports BIG-Bench Hard 26.7 | 268M | 32K | — | Laptop / edge | Gemma Terms of Use attribution | 0 | — |
| Kimi-Linear-48B-A3B-Instruct moonshotai | author-reported Author reports MMLU-Pro (4k context length) 51.0 | 49.1B (3B active) | — | — | Multi-GPU node | MIT | 1 | — |
| Qwen3-30B-A3B-Thinking-2507 Qwen | author-reported Author reports MMLU-Pro 80.9Author reports MMLU-Redux 91.4 +1 | 30.5B (3B active) | 256K | — | Single high-end GPU | Apache 2.0 | 1 | — |
| gemma-4-E2B google | author-reported Author reports MMLU Pro 60.0%Author reports AIME 2026 no tools 37.5% +1 | 5.12B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| Qwen2-0.5B-Instruct Qwen | author-reported Author reports MMLU 37.9Author reports GSM8K 40.1 | 494M | 32K | — | Laptop / edge | Apache 2.0 | 1 | — |
| Qwen2-0.5B Qwen | author-reported Author reports MMLU 45.4Author reports MMLU-Pro 14.7 +1 | 494M | 128K | — | Laptop / edge | Apache 2.0 | 1 | — |
| Qwen3-Next-80B-A3B-Thinking Qwen | author-reported Author reports MMLU-Pro 82.7Author reports MMLU-Redux 92.5 +1 | 81.3B (3B active) | 256K | $0.15 | Multi-GPU node | Apache 2.0 | 1 | — |
| gemma-4-12B google | author-reported Author reports MMLU Pro 77.2%Author reports AIME 2026 no tools 77.5% +1 | 12.0B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| chandra-ocr-2 datalab-to | author-reported Author reports old_scans_math 89.1chat template: <think> | 5.30B | 256K | — | Single GPU | OpenRAIL attribution | 0 | — |
| gpt-neox-20b EleutherAI | author-reported Author reports MMLU (5-shot) 25.0Author reports GSM8K (5-shot) 2.43 | 20.7B | 2K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| Qianfan-OCR baidu | author-reported Author reports arxiv_math 80.1chat template: <think> | 4.74B | 32K | — | Single GPU | Apache 2.0 | 0 | — |
| Qwen-7B-Chat Qwen | author-reported Author reports MMLU 55.8Author reports GSM8K 50.3 | 7.72B | 32K | — | Single GPU | tongyi-qianwen-license-ag… review | 0 | — |
| gemma-4-31B-it-assistant google | author-reported Author reports MMLU Pro 85.2%Author reports AIME 2026 no tools 89.2% +1 | 470M | 256K | — | Laptop / edge | Apache 2.0 | 0 | — |
| surya-ocr-2 datalab-to | author-reported Author reports arxiv_math 88.3 | 686M | 256K | — | Laptop / edge | OpenRAIL attribution | 1 | — |
| DeepSeek-V2-Lite deepseek-ai | author-reported Author reports MMLU 58.3Author reports BBH 48.1 +1 | 15.7B | 160K | — | Single high-end GPU | DeepSeek License (deepsee… attribution | 0 | — |
| Inkling-Small thinkingmachines | author-reported Author reports MathArena/aime_2026 95.5Author reports hle 31.6 +1 | 266B | 1M | $0.00 | Multi-GPU node | Apache 2.0 | 2 | — |
| gemma-3-4b-pt google | author-reported Author reports BIG-Bench Hard 50.9Author reports MMLU 59.6 +1 | 4.30B | — | — | Single GPU | Gemma Terms of Use attribution | 1 | — |
| gpt-neo-125m EleutherAI | author-reported Author reports MMLU (5-shot) 25.97Author reports GSM8K (5-shot) 0.3 | 150M | 2K | — | Laptop / edge | MIT | 0 | — |
| Phi-3.5-MoE-instruct microsoft | author-reported Author reports BigBench Hard CoT (0-shot) 79.1Author reports MMLU (5-shot) 78.9 +1 | 41.9B | 128K | — | Multi-GPU node | MIT | 0 | — |
| gpt-neo-2.7B EleutherAI | author-reported Author reports MathQA 24.72% | 2.72B | 2K | — | Single GPU | MIT | 0 | — |
| DeepSeek-V4-Flash-DSpark deepseek-ai | author-reported Author reports MMLU-Redux (EM) 89.4Author reports MMLU-Pro (EM) 87.5 +1 | 165B | 1M | — | Multi-GPU node | MIT | 0 | — |
| LFM2.5-8B-A1B LiquidAI | author-reported Author reports MathArena/aime_2026 50.0chat template: <think> | 8.47B (1B active) | 125K | — | Single GPU | lfm1.0 review | 0 | — |
| MiMo-V2.5-Pro XiaomiMiMo | author-reported Author reports BBH 88.4Author reports MMLU-Redux 92.8 +1 | 1.02T | 1M | $0.30 | Cluster | MIT | 2 | — |
| OpenHermes-2.5-Mistral-7B teknium | author-reported Author reports agieval_sat_math 0.3773 | 7.24B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen2-7B Qwen | author-reported Author reports MMLU 70.3Author reports MMLU-Pro 40.0 +1 | 7.62B | 128K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen-72B Qwen | author-reported Author reports MMLU 77.4Author reports GSM8K 78.9 +1 | 72.3B | 32K | — | Multi-GPU node | tongyi-qianwen-license-ag… review | 0 | — |
| Qwen-7B Qwen | author-reported Author reports MMLU 58.2Author reports GSM8K 51.7 +1 | 7.72B | 32K | — | Single GPU | tongyi-qianwen-license-ag… review | 0 | — |
| Ling-3.0-flash inclusionAI | author-reported Author reports MathArena/aime_2026 93.2Author reports hle 22.7 +1 | 127B | 256K | $0.02 | Multi-GPU node | MIT | 3 | — |
| DeepSeek-V2-Lite-Chat deepseek-ai | author-reported Author reports MMLU 58.3Author reports BBH 48.1 +1 | 15.7B | 160K | — | Single high-end GPU | DeepSeek License (deepsee… attribution | 0 | — |
| gemma-4-26B-A4B-it-assistant google | author-reported Author reports MMLU Pro 82.6%Author reports AIME 2026 no tools 88.3% +1 | 420M (4B active) | 256K | — | Laptop / edge | Apache 2.0 | 0 | — |
| gemma-3-1b-pt google | author-reported Author reports BIG-Bench Hard 28.4 | 1000M | 32K | — | Laptop / edge | Gemma Terms of Use attribution | 1 | — |
| gpt-neo-1.3B EleutherAI | author-reported Author reports MathQA 24.05%Author reports MMLU (5-shot) 24.82 +1 | 1.37B | 2K | — | Laptop / edge | MIT | 0 | — |
| granite-4.1-8b ibm-granite | author-reported Author reports MMLU 73.84Author reports MMLU-Pro 55.99 +1 | 8.79B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| Qwen2-1.5B Qwen | author-reported Author reports MMLU 56.5Author reports MMLU-Pro 21.8 +1 | 1.54B | 128K | — | Laptop / edge | Apache 2.0 | 1 | — |
| bitnet-b1.58-2B-4T microsoft | author-reported Author reports MMLU 53.17 | 850M | 4K | — | Laptop / edge | MIT | 0 | — |
| DialoGPT-small microsoft | author-reported Author reports MMLU (5-shot) 25.81Author reports GSM8K (5-shot) 0.0 | 176M | 1K | — | Laptop / edge | MIT | 0 | — |
| granite-4.1-3b ibm-granite | author-reported Author reports MMLU 67.02Author reports MMLU-Pro 49.83 +1 | 3.40B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| EXAONE-3.5-7.8B-Instruct LGAI-EXAONE | author-reported Author reports LiveBench 39.8 | 7.82B | 32K | — | Single GPU | exaone review | 0 | — |
| Hunyuan-A13B-Instruct tencent | author-reported Author reports MMLU 88.17Author reports MMLU-Pro 67.23 +1 | 80.4B (13B active) | 32K | $0.14 | Multi-GPU node | tencent-hunyuan-a13b review | 1 | — |
| openchat_3.5 openchat | author-reported Author reports MMLU 64.3Author reports MATH 28.6 +1 | — | 8K | — | — | Apache 2.0 | 2 | — |
| DeepSeek-V3.1-Base deepseek-ai | author-reported Author reports MMLU-Redux (EM) 91.8Author reports MMLU-Pro (EM) 83.7 +1 | 685B | 160K | — | Cluster | MIT | 0 | — |
| Phi-4-mini-reasoning microsoft | author-reported Author reports AIME 57.5Author reports MATH-500 94.6 +1 | 3.84B | 128K | — | Single GPU | MIT | 1 | — |
| Olmo-3-7B-Instruct allenai | author-reported Author reports MATH 87.3Author reports AIME 2024 44.3 +1 | 7.30B | 64K | — | Single GPU | Apache 2.0 | 0 | — |
| LFM2.5-350M LiquidAI | author-reported Author reports mmlu_pro 20.01chat template: <think> | 354M | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| gemma-3-12b-pt google | author-reported Author reports BIG-Bench Hard 72.6Author reports MMLU 74.5 +1 | 12.2B | — | — | Single high-end GPU | Gemma Terms of Use attribution | 1 | — |
| Llama-2-7b-hf NousResearch | author-reported Author reports Math 14.6Author reports MMLU 45.3 +1 | 6.74B | 4K | — | Single GPU | Not declared review | 1 | — |
| OLMo-2-0425-1B allenai | author-reported Author reports BBH 35.0Author reports GSM8K 68.3 +1 | 1.48B | 4K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Jamba-v0.1 ai21labs | author-reported Author reports MMLU 67.4%Author reports BBH 45.4% +1 | 51.6B | 256K | — | Multi-GPU node | Apache 2.0 | 0 | — |
| NVLM-D-72B nvidia | author-reported Author reports MMLU 82.0Author reports GSM8K 93.2 +1 | 79.4B | — | — | Multi-GPU node | CC BY-NC 4.0 non-commercial | 0 | — |
| DeepSeek-V2-Chat deepseek-ai | author-reported Author reports MMLU 78.5Author reports BBH 81.3 +1 | 236B | 160K | — | Multi-GPU node | DeepSeek License (deepsee… attribution | 0 | — |
| Llama-3.1-Nemotron-Nano-8B-v1 nvidia | author-reported Author reports MATH500 95.4%Author reports AIME25 47.1% +1 | 8.03B | 128K | — | Single GPU | NVIDIA Open Model License… attribution | 1 | — |
| Meta-Llama-3.1-8B-Instruct NousResearch | author-reported Author reports MMLU 69.4Author reports MMLU-Pro (CoT) 37.1 +1 | 8.03B | 128K | — | Single GPU | Llama 3.1 Community Licen… attribution | 1 | — |
| t5gemma-2-270m-270m google | author-reported Author reports BIG-Bench Hard 22.9Author reports MMLU (Pro CoT) 10.6 +1 | 786M | — | — | Laptop / edge | Gemma Terms of Use attribution | 0 | — |
| MiMo-V2-Flash XiaomiMiMo | author-reported Author reports BBH 88.5Author reports MMLU-Redux 90.6 +1 | 310B | 256K | — | Cluster | MIT | 0 | — |
| Nanbeige4.2-3B Nanbeige | author-reported Author reports GPQA-Diamond 87.4Author reports HMMT-Feb-2026 82.8 +1 | 4.17B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| LFM2-1.2B LiquidAI | author-reported Author reports MMLU 55.23Author reports GPQA 31.47 +1 | 1.17B | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| Ornith-1.5-9B ornith-ai | author-reported Author reports GPQA Diamond 86.4chat template: <think> | 9.65B | 256K | — | Single GPU | MIT | 1 | — |
| LFM2.5-230M LiquidAI | author-reported Author reports mmlu_pro 20.25chat template: <think> | 230M | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| gemma-3n-E2B-it-litert-lm google | author-reported Author reports BIG-Bench Hard 44.3 | — | — | — | — | Gemma Terms of Use attribution | 0 | — |
| Qwen2-72B Qwen | author-reported Author reports MMLU 84.2Author reports MMLU-Pro 55.6 +1 | 72.7B | 128K | — | Multi-GPU node | tongyi-qianwen review | 1 | — |
| Kimi-K3-DSpark RadixArk | author-reported Author reports GSM8K 5.4176Author reports MATH500 4.1329 +1 | 2.25B | 1M | — | Single GPU | Not declared review | 0 | — |
| Llama-3_3-Nemotron-Super-49B-v1 nvidia | author-reported Author reports MATH500 96.6Author reports AIME25 58.4 +1 | 49.9B | 128K | — | Multi-GPU node | NVIDIA Open Model License… attribution | 0 | — |
| Moonlight-16B-A3B-Instruct moonshotai | author-reported Author reports MMLU 70.0Author reports MMLU-pro 42.4 +1 | 16.0B (3B active) | 8K | — | Single high-end GPU | MIT | 0 | — |
| gemma-2-27b google | author-reported Author reports MMLU 75.2Author reports GSM8K 74.0 | 27.2B | 8K | $0.65 | Single high-end GPU | Gemma Terms of Use attribution | 1 | — |
| Nemotron-Cascade-2-30B-A3B nvidia | author-reported Author reports IMO AnswerBench 79.3Author reports AIME 2025 92.4 (98.6)† +1 | 31.6B (3B active) | 256K | — | Single high-end GPU | NVIDIA Open Model License… attribution | 0 | — |
| granite-4.0-h-tiny ibm-granite | author-reported Author reports MMLU 68.65Author reports MMLU-Pro 44.94 +1 | 6.94B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| OpenELM-1_1B-Instruct apple | author-reported Author reports MMLU 25.65 | 1.08B | — | — | Laptop / edge | apple-sample-code-license review | 0 | — |
| Step3-VL-10B stepfun-ai | author-reported Author reports AIME 2025 94.43Author reports HMMT 2025 92.14 +1 | 10.2B | 64K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| Llama-2-7b-chat-hf NousResearch | author-reported Author reports Math 14.6Author reports MMLU 45.3 +1 | 6.74B | 4K | — | Single GPU | Not declared review | 1 | — |
| Nanbeige4.1-3B Nanbeige | author-reported Author reports AIME 2026 I 87.40Author reports HMMT Nov 77.92 +1 | 3.93B | 256K | — | Single GPU | Apache 2.0 | 1 | — |
| Solar-Open-100B upstage | author-reported Author reports KMMLU 73.0chat template: <|think|> | 103B | 128K | — | Multi-GPU node | upstage-solar-license review | 0 | — |
| SmolLM2-360M HuggingFaceTB | author-reported Author reports MMLU (cloze) 35.8Author reports GSM8K (5-shot) 7.43 | 362M | 8K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Olmo-3-7B-Think allenai | author-reported Author reports MATH 95.1Author reports AIME 2024 74.6 +1 | 7.30B | 64K | — | Single GPU | Apache 2.0 | 0 | — |
| internlm3-8b-instruct internlm | author-reported Author reports MMLU(0-shot) 76.6Author reports MMLU-Pro(0-shot) 57.6 +1 | 8.80B | 32K | — | Single GPU | Apache 2.0 | 0 | — |
| EXAONE-4.5-33B LGAI-EXAONE | author-reported Author reports WeMath 79.1chat template: <think> | 34.4B | 256K | — | Multi-GPU node | exaone review | 0 | — |
| Hy3-preview tencent | author-reported Author reports MMLU 87.42Author reports MMLU-Pro 65.76 +1 | 299B | 256K | $0.18 | Cluster | Other review | 1 | — |
| granite-4.0-h-small ibm-granite | author-reported Author reports MMLU 78.44Author reports MMLU-Pro 55.47 +1 | 32.2B | 128K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| DeepSeek-V2 deepseek-ai | author-reported Author reports MMLU 78.5Author reports BBH 81.3 +1 | 236B | 160K | — | Multi-GPU node | DeepSeek License (deepsee… attribution | 0 | — |
| EXAONE-3.5-2.4B-Instruct LGAI-EXAONE | author-reported Author reports LiveBench 33.0 | 2.41B | 32K | — | Single GPU | exaone review | 0 | — |
| gemma-3n-E4B-it-litert-lm google | author-reported Author reports BIG-Bench Hard 52.9 | — | — | — | — | Gemma Terms of Use attribution | 0 | — |
| Kimi-VL-A3B-Thinking-2506 moonshotai | author-reported Author reports MMLU 82.0Author reports MMLU-Pro 68.5 +1 | 16.4B (3B active) | 128K | — | Single high-end GPU | MIT | 0 | — |
| OLMoE-1B-7B-0924 allenai | author-reported Author reports MMLU 54.1 | 6.92B | 4K | — | Single GPU | Apache 2.0 | 0 | — |
| K-EXAONE-236B-A23B LGAI-EXAONE | author-reported Author reports MMLU-Pro 83.8Author reports GPQA-Diamond 79.1 +1 | 237B (23B active) | 256K | — | Multi-GPU node | k-exaone review | 0 | — |
| Llama-2-13b-hf NousResearch | author-reported Author reports Math 28.7Author reports MMLU 54.8 +1 | 13.0B | 4K | — | Single high-end GPU | Not declared review | 1 | — |
| Phi-tiny-MoE-instruct microsoft | author-reported Author reports MMLU 60.83Author reports MMLU pro 36.34 +1 | 3.76B | 4K | — | Single GPU | MIT | 0 | — |
| kev-4b jaredpalmer | author-reported Author reports MMLU-Pro, 10 options (transfer-v9 development) 0.565 | — | — | — | — | Apache 2.0 | 0 | — |
| pplx-decider-v1-27b perplexity-ai | author-reported Author reports BBH 82.80%chat template: <think> | 26.1B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| kev-9b jaredpalmer | author-reported Author reports MMLU-Pro, 10 options (transfer-v9 development) 0.590chat template: <think> | — | — | — | — | Apache 2.0 | 0 | — |
| kev-0.8b jaredpalmer | author-reported Author reports MMLU-Pro, 10 options (transfer-v9 development) 0.230 | — | — | — | — | Apache 2.0 | 0 | — |
| jpt-4b kirp | author-reported Author reports Knowledge GPQA Diamond 23.8Author reports Knowledge GSM8K 66.6 +1 | 4.54B | 256K | — | Single GPU | CC BY-NC 4.0 non-commercial | 0 | — |
| jpt-9b kirp | author-reported Author reports BBH 55.6Author reports GPQA Diamond 21.8 +1 | 9.41B | 256K | — | Single GPU | CC BY-NC 4.0 non-commercial | 0 | — |
| Qwen3.8-27B Qwen | structural decision-model estimate p=1.00chat template: <think> | 27.8B | 256K | $0.02 | Single high-end GPU | Apache 2.0 | 7 | 28.6 s · 87 tok/s |
| gpt-oss-120b openai | structural decision-model estimate p=0.99chat template: <|channel|>analysis | 117B | 128K | $0.03 | Multi-GPU node | Apache 2.0 | 13 | 0.8 s · 168 tok/s |
| gpt-oss-20b openai | structural decision-model estimate p=0.96chat template: <|channel|>analysis | 20.9B | 128K | $0.02 | Single high-end GPU | Apache 2.0 | 7 | 2.8 s · 58 tok/s |
| Qwen3.6-35B-A3B Qwen | structural chat template: <think> | 36.0B (3B active) | 256K | $0.05 | Multi-GPU node | Apache 2.0 | 5 | — |
| Qwen3-VL-8B-Instruct Qwen | structural decision-model estimate p=0.72 | 8.77B | 256K | $0.25 | Single GPU | Apache 2.0 | 4 | — |
| Qwen3-8B Qwen | structural decision-model estimate p=0.93chat template: <think> | 8.19B | 40K | $0.07 | Single GPU | Apache 2.0 | 3 | — |
| MiniMax-M3 MiniMaxAI | structural chat template: reasoning_content | 427B | 1M | $0.22 | Cluster | MiniMax Model License (mi… gated-threshold | 7 | — |
| Qwen3-14B Qwen | structural decision-model estimate p=0.82chat template: <think> | 14.8B | 40K | $0.07 | Single high-end GPU | Apache 2.0 | 4 | — |
| Qwen3-0.6B Qwen | structural decision-model estimate p=0.78chat template: <think> | 752M | 40K | — | Laptop / edge | Apache 2.0 | 1 | — |
| Qwen3-32B Qwen | structural decision-model estimate p=0.87chat template: <think> | 32.8B | 40K | $0.08 | Single high-end GPU | Apache 2.0 | 4 | — |
| Qwen3.6-27B Qwen | structural chat template: <think> | 27.8B | 256K | $0.20 | Single high-end GPU | Apache 2.0 | 4 | — |
| Qwen3.8-Flash-Next Qwen | structural chat template: <think> | 180B | 256K | $0.15 | Multi-GPU node | Qwen License (qwen-commun… gated-threshold | 2 | — |
| Qwen3-235B-A22B Qwen | structural decision-model estimate p=0.94chat template: <think> | 235B (22B active) | 40K | $0.20 | Multi-GPU node | Apache 2.0 | 4 | — |
| Qwen3-Embedding-0.6B Qwen | structural chat template: <think> | 596M | 32K | $0.01 | Laptop / edge | Apache 2.0 | 1 | — |
| Kimi-K2-Instruct moonshotai | structural decision-model estimate p=0.88 | 1.03T | 128K | $0.57 | Cluster | Kimi (Modified MIT) (modi… gated-threshold | 4 | — |
| DeepSeek-V3-0324 deepseek-ai | structural decision-model estimate p=0.95 | 685B | 160K | $0.24 | Cluster | MIT | 3 | — |
| GLM-4.5-Air zai-org | structural decision-model estimate p=0.98chat template: <think> | 110B | 128K | $0.13 | Multi-GPU node | MIT | 4 | — |
| MiniMax-M2.7 MiniMaxAI | structural chat template: <think> | 229B | 200K | $0.21 | Multi-GPU node | other review | 4 | — |
| Qwen3-4B Qwen | structural decision-model estimate p=0.76chat template: <think> | 4.02B | 40K | — | Single GPU | Apache 2.0 | 1 | — |
| DeepSeek-V3.2 deepseek-ai | structural decision-model estimate p=0.96 | 685B | 160K | $0.21 | Cluster | MIT | 2 | — |
| Qwen3-VL-4B-Instruct Qwen | structural decision-model estimate p=0.70 | 4.44B | 256K | — | Single GPU | Apache 2.0 | 2 | — |
| Qwen3-30B-A3B Qwen | structural decision-model estimate p=0.89chat template: <think> | 30.5B (3B active) | 40K | $0.12 | Single high-end GPU | Apache 2.0 | 2 | — |
| Qwen3-1.7B Qwen | structural decision-model estimate p=0.77chat template: <think> | 2.03B | 40K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen3-Reranker-0.6B Qwen | structural chat template: <think> | 596M | 40K | $0.01 | Laptop / edge | Apache 2.0 | 2 | — |
| GLM-4.6 zai-org | structural decision-model estimate p=0.73chat template: <think> | 357B | 198K | $0.40 | Cluster | MIT | 5 | — |
| MiniMax-M2 MiniMaxAI | structural chat template: <think> | 229B | 192K | $0.26 | Multi-GPU node | MiniMax (Modified MIT) (m… review | 3 | — |
| Qwen3-VL-235B-A22B-Instruct Qwen | structural decision-model estimate p=0.73 | 236B (22B active) | 256K | $0.20 | Multi-GPU node | Apache 2.0 | 3 | — |
| Kimi-K2.7-Code moonshotai | structural chat template: <think> | 1.03T | 256K | $0.67 | Cluster | Kimi (Modified MIT) (modi… gated-threshold | 4 | — |
| Qwen3.8-2.4T-A95B Qwen | structural decision-model estimate p=0.99chat template: <think> | 2.45T (95B active) | 256K | $2.00 | Cluster | Qwen License (qwen3.8-max) gated-threshold | 5 | — |
| Qwen3-Reranker-4B Qwen | structural chat template: <think> | 4.02B | 40K | $0.03 | Single GPU | Apache 2.0 | 1 | — |
| QwQ-32B Qwen | structural decision-model estimate p=1.00chat template: <think> | 32.8B | 40K | — | Single high-end GPU | Apache 2.0 | 2 | — |
| Hy3 tencent | structural chat template: reasoning_content | 299B | 256K | $0.08 | Cluster | Apache 2.0 | 3 | — |
| GLM-4.5 zai-org | structural decision-model estimate p=0.99chat template: <think> | 358B | 128K | $0.60 | Cluster | MIT | 2 | — |
| MiniMax-M2.1 MiniMaxAI | structural chat template: <think> | 229B | 192K | $0.30 | Multi-GPU node | MiniMax (Modified MIT) (m… review | 3 | — |
| Qwen3-VL-235B-A22B-Thinking Qwen | structural decision-model estimate p=1.00name: Thinking +1 | 236B (22B active) | 256K | — | Multi-GPU node | Apache 2.0 | 3 | — |
| GLM-4.5V zai-org | structural decision-model estimate p=0.81chat template: <think> | 108B | 64K | $0.60 | Multi-GPU node | MIT | 3 | — |
| Phi-3.5-mini-instruct microsoft | structural decision-model estimate p=0.87 | 3.82B | 128K | — | Single GPU | MIT | 0 | — |
| Qwen3-0.6B-Base Qwen | structural chat template: <think> | 596M | 32K | — | Laptop / edge | Apache 2.0 | 1 | — |
| Qwen3-4B-Base Qwen | structural chat template: <think> | 4.02B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen3-Reranker-8B Qwen | structural chat template: <think> | 8.19B | 40K | $0.05 | Single GPU | Apache 2.0 | 1 | — |
| Qwen3-8B-Base Qwen | structural chat template: <think> | 8.19B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen3-VL-8B-Thinking Qwen | structural decision-model estimate p=1.00name: Thinking +1 | 8.77B | 256K | — | Single GPU | Apache 2.0 | 3 | — |
| QwQ-32B-Preview Qwen | structural decision-model estimate p=1.00 | 32.8B | 32K | — | Single high-end GPU | Apache 2.0 | 2 | — |
| Apertus-8B-Instruct-2509 swiss-ai | structural chat template: enable_thinking | 8.05B | 64K | $0.10 | Single GPU | Apache 2.0 | 2 | — |
| GLM-4.6V-Flash zai-org | structural chat template: <think> | 10.3B | 128K | $0.30 | Single high-end GPU | MIT | 2 | — |
| Qwen3-1.7B-Base Qwen | structural chat template: <think> | 1.72B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen2.5-Math-1.5B Qwen | structural decision-model estimate p=1.00name: Math | 1.54B | 4K | — | Laptop / edge | Apache 2.0 | 1 | — |
| LFM2.5-2.6B LiquidAI | structural chat template: <think> | 2.70B | 128K | $0.00 | Single GPU | lfm1.0 review | 1 | — |
| granite-docling-258M ibm-granite | structural tag: math | 258M | 8K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Qwen3Guard-Gen-0.6B Qwen | structural chat template: <think> | 752M | 32K | — | Laptop / edge | Apache 2.0 | 1 | — |
| GLM-4.1V-9B-Thinking zai-org | structural decision-model estimate p=1.00name: Thinking +1 | 10.3B | 64K | — | Single high-end GPU | MIT | 0 | — |
| MiMo-V2.6-Pro-RL XiaomiMiMo | structural chat template: <think> | 1.02T | 1M | $0.39 | Cluster | MIT | 2 | — |
| LFM2.5-1.2B-Instruct LiquidAI | structural chat template: <think> | 1.17B | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| MiniCPM5-2B openbmb | structural chat template: <think> | 2.52B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| MiMo-V2.6-Flash-RL XiaomiMiMo | structural chat template: <think> | 311B | 1M | $0.09 | Cluster | MIT | 2 | — |
| Hy4-preview tencent | structural chat template: reasoning_content | 780B | 1M | $0.75 | Cluster | Apache 2.0 | 3 | — |
| medgemma-27b-text-it google | structural decision-model estimate p=0.87tag: thinking | 27.0B | — | — | Single high-end GPU | health-ai-developer-found… review | 1 | — |
| Laguna-S-2.1 poolside | structural decision-model estimate p=0.84chat template: <think> | 118B | 1M | $0.00 | Multi-GPU node | — review | 2 | — |
| Qwen3-VL-4B-Thinking Qwen | structural decision-model estimate p=0.99name: Thinking +1 | 4.44B | 256K | — | Single GPU | Apache 2.0 | 2 | — |
| Step-3.7-Flash stepfun-ai | structural decision-model estimate p=0.76chat template: <think> | 201B | 256K | $0.20 | Multi-GPU node | Apache 2.0 | 3 | — |
| Ornith-1.0-9B ornith-ai | structural chat template: <think> | 1.47M | 256K | — | Laptop / edge | MIT | 1 | — |
| Qwen3-VL-30B-A3B-Thinking Qwen | structural decision-model estimate p=1.00name: Thinking +1 | 31.1B (3B active) | 256K | $0.29 | Single high-end GPU | Apache 2.0 | 2 | — |
| GLM-4.6V zai-org | structural chat template: <think> | 108B | 128K | $0.30 | Multi-GPU node | MIT | 3 | — |
| gpt-oss-safeguard-20b openai | structural chat template: <|channel|>analysis | 21.5B | 128K | $0.07 | Single high-end GPU | Apache 2.0 | 2 | — |
| MiniCPM-o-4_5 openbmb | structural chat template: <think> | 9.37B | 40K | — | Single GPU | Apache 2.0 | 0 | — |
| MiMo-V2.5 XiaomiMiMo | structural chat template: <think> | 311B | 1M | $0.12 | Cluster | MIT | 2 | — |
| Qwen3Guard-Gen-8B Qwen | structural chat template: <think> | 8.19B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen2.5-Math-1.5B-Instruct Qwen | structural decision-model estimate p=1.00name: Math | 1.54B | 4K | — | Laptop / edge | Apache 2.0 | 1 | — |
| MiniCPM5-1B openbmb | structural decision-model estimate p=0.97chat template: <think> | 1.08B | 128K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Qwen3-14B-Base Qwen | structural chat template: <think> | 14.8B | 32K | — | Single high-end GPU | Apache 2.0 | 1 | — |
| Qwen2.5-Math-7B Qwen | structural decision-model estimate p=1.00name: Math | 7.62B | 4K | — | Single GPU | Apache 2.0 | 1 | — |
| Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled Jackrong | structural decision-model estimate p=1.00name: Reasoning +1 | 27.8B | 256K | — | Single high-end GPU | Apache 2.0 | 1 | — |
| QVQ-72B-Preview Qwen | structural decision-model estimate p=0.92 | 73.4B | 125K | — | Multi-GPU node | Qwen License (qwen) gated-threshold | 1 | — |
| Qwen3.5-0.8B-Base Qwen | structural chat template: <think> | 873M | 256K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Qwen2.5-Math-7B-Instruct Qwen | structural decision-model estimate p=1.00name: Math | 7.62B | 4K | — | Single GPU | Apache 2.0 | 1 | — |
| MiniCPM-V-4.6 openbmb | structural chat template: <think> | 1.30B | 256K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Ornith-1.0-35B ornith-ai | structural chat template: <think> | 665K | 256K | — | Laptop / edge | MIT | 0 | — |
| OvisOCR2 ATH-MaaS | structural chat template: <think> | 853M | 256K | — | Laptop / edge | Apache 2.0 | 1 | — |
| Qwen3-Omni-30B-A3B-Thinking Qwen | structural name: Thinking | 31.7B (3B active) | — | — | Single high-end GPU | Apache 2.0 (apache-2.0) | 0 | — |
| Qwen3.5-4B-Base Qwen | structural chat template: <think> | 4.66B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Qwen3.5-9B-Base Qwen | structural chat template: <think> | 9.65B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Llama-3_3-Nemotron-Super-49B-v1_5 nvidia | structural decision-model estimate p=0.99chat template: <think> | 49.9B | 128K | — | Multi-GPU node | NVIDIA Open Model License… attribution | 0 | — |
| Qwen-AgentWorld-35B-A3B Qwen | structural chat template: <think> | 34.7B (3B active) | 256K | — | Multi-GPU node | Apache 2.0 | 0 | — |
| Orca-2-13b microsoft | structural decision-model estimate p=0.93 | — | 4K | — | — | Research / non-commercial… non-commercial | 1 | — |
| Qwen3.5-2B-Base Qwen | structural chat template: <think> | 2.27B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| DeepSeek-R1-Zero deepseek-ai | structural decision-model estimate p=1.00name: R1 +1 | 684B | 160K | — | Cluster | MIT | 0 | — |
| MiniCPM-V-4_5 openbmb | structural chat template: <think> | 8.70B | 40K | — | Single GPU | Apache 2.0 | 0 | — |
| GLM-4-32B-0414 zai-org | structural decision-model estimate p=0.71 | 32.6B | 32K | — | Single high-end GPU | MIT | 1 | — |
| granite-4.2-8b ibm-granite | structural decision-model estimate p=1.00tag: reasoning +1 | 8.79B | 128K | $0.06 | Single GPU | Apache 2.0 | 2 | — |
| ERNIE-4.5-21B-A3B-Thinking baidu | structural decision-model estimate p=1.00name: Thinking +1 | 21.8B (3B active) | 128K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| LFM2.5-VL-450M LiquidAI | structural chat template: <think> | 449M | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| Qwen3Guard-Gen-4B Qwen | structural chat template: <think> | 4.41B | 32K | — | Single GPU | Apache 2.0 | 1 | — |
| Kimi-VL-A3B-Thinking moonshotai | structural decision-model estimate p=1.00name: Thinking | 16.4B (3B active) | 128K | — | Single high-end GPU | MIT | 0 | — |
| North-Mini-Code-1.0 CohereLabs | structural chat template: reasoning_content | 30.5B | 488K | $0.00 | Single high-end GPU | Apache 2.0 | 1 | — |
| Phi-4-reasoning-plus microsoft | structural decision-model estimate p=1.00name: reasoning +1 | 14.7B | 32K | — | Single high-end GPU | MIT | 1 | — |
| Qwen3-VL-2B-Thinking Qwen | structural decision-model estimate p=0.99name: Thinking +1 | 2.13B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Ornith-1.5-35B-A3B ornith-ai | structural chat template: <think> | 36.0B (3B active) | 256K | — | Multi-GPU node | MIT | 0 | — |
| LFM2.5-VL-1.6B LiquidAI | structural chat template: thinking | 1.60B | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| MiMo-7B-RL XiaomiMiMo | structural decision-model estimate p=1.00 | 7.83B | 32K | — | Single GPU | MIT | 0 | — |
| granite-3.3-8b-instruct ibm-granite | structural decision-model estimate p=0.96chat template: <think> | 8.17B | 128K | — | Single GPU | Apache 2.0 | 0 | — |
| Cosmos-Reason1-7B nvidia | structural decision-model estimate p=0.86 | 8.29B | 125K | — | Single GPU | NVIDIA Open Model License… attribution | 0 | — |
| Phi-4-reasoning microsoft | structural decision-model estimate p=0.96name: reasoning +1 | 14.7B | — | — | Single high-end GPU | MIT | 0 | — |
| Nemotron-Research-Reasoning-Qwen-1.5B nvidia | structural decision-model estimate p=0.99name: Reasoning | 1.78B | — | — | Single GPU | CC BY-NC 4.0 non-commercial | 1 | — |
| Intern-Decision-4B internlm | structural chat template: <think> | 4.54B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Intern-Decision-0.8B internlm | structural chat template: <think> | 853M | 256K | — | Laptop / edge | Apache 2.0 | 0 | — |
| Winnow-12B EldanRing | structural chat template: <|think|> | — | 256K | — | — | Apache 2.0 | 0 | — |
| JevK5 alibiserikbay | structural chat template: <think> | 4.21B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Intern-Decision-2B internlm | structural chat template: <think> | 2.21B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Tev1-4B-experimental togethercomputer | structural chat template: <think> | 4.66B | 256K | $0.04 | Single GPU | Not declared review | 1 | — |
| Tev1-0.8B-experimental togethercomputer | structural chat template: <think> | 873M | 256K | — | Laptop / edge | Not declared review | 0 | — |
| xor juspay | structural chat template: <think> | 35.1B | 256K | — | Multi-GPU node | Apache 2.0 | 0 | — |
| LFM2.5-350M-RLCD notnotsamuel | structural chat template: <think> | 354M | 125K | — | Laptop / edge | lfm1.0 review | 0 | — |
| openjev openjev | structural chat template: <think> | 27.4B | 256K | — | Single high-end GPU | CC BY-NC 4.0 non-commercial | 0 | — |
| Jev-Omni akhilaaa3 | structural chat template: <|think|> | 12.0B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| JEV-27B autotrust | structural chat template: <think> | 26.9B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| Bespoke-Nimble-9B bespokelabs | structural chat template: <think> | — | — | — | — | Apache 2.0 | 0 | — |
| lev interfaze-ai | structural chat template: <think> | — | — | — | — | Apache 2.0 | 0 | — |
| decider-2b Mapika | structural chat template: <think> | 1.88B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| decider-4b Mapika | structural chat template: <think> | 4.21B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| LFM2.5-2.6B-RLCD monotykamary | structural chat template: <think> | 2.70B | 128K | — | Single GPU | lfm1.0 review | 0 | — |
| system-one-qwen3.5-4b-scorer-v2b pngwn | structural chat template: <think> | — | — | — | — | CC BY-NC 4.0 non-commercial | 0 | — |
| this-that-model-1.0 flock-io | structural chat template: <think> | 1.88B | 256K | — | Single GPU | MIT | 0 | — |
| decider-35b-a3b-nvfp4 Mapika | structural chat template: <think> | 17.9B (3B active) | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
| metask-jev-4b-policy-mix wayfind | structural chat template: <think> | 4.54B | 256K | — | Single GPU | Apache 2.0 | 0 | — |
| Bespoke-Nimble-9B-v2 bespokelabs | structural chat template: <think> | — | — | — | — | Apache 2.0 | 0 | — |
| this-that-model-1.2 flock-io | structural chat template: <think> | 1.88B | 256K | — | Single GPU | MIT | 0 | — |
| Eikos-27B-FP8 caiovicentino1 | structural chat template: <think> | 27.8B | 256K | — | Single high-end GPU | MIT | 0 | — |
| blink-27b thegovind | structural chat template: <think> | 26.9B | 256K | — | Single high-end GPU | Research / non-commercial… non-commercial | 0 | — |
| jpt-0.8b kirp | structural chat template: <think> | 853M | 256K | — | Laptop / edge | CC BY-NC 4.0 non-commercial | 0 | — |
| jebadiah-27b frontier-infra | structural chat template: <think> | 27.8B | 256K | — | Single high-end GPU | Apache 2.0 | 0 | — |
No model matches every constraint. Loosen one, or switch the evidence filter back to any.
How to read this
- measured a third-party leaderboard row for this model. Settings (agent harness, edit format, prompt) differ between boards; compare within a benchmark, not across.
- author-reported a number the model author published in the card or eval files. Not reproduced by us.
- structural a deterministic signal from the name, config or chat template, or a decision-model estimate. A weak signal: it says the model was built or tagged for the job, not how well it does it. "decision-model estimate" chips are a probability from the Jev decision model, not a measurement.
- Runs on is the memory class of the BF16 weights (method); choosing a class shows that class and smaller. Quantized builds may fit a class lower.
- Price from is the cheapest observed hosted input price, USD per million tokens; models with no observed price are hidden when a ceiling is set. License friction: permissive = none or attribution-only; high = non-commercial.
- Ordering within a tier is the watch-list score (Hub likes, downloads, recency), not a quality verdict.
Subscribe
Follow "Reason through hard problems"
- /feeds/for/reasoning-math.xml(RSS 2.0) · one item per model with measured or author-reported evidence for this job
- /feeds/for/reasoning-math.json(JSON Feed 1.1) · same items, JSON Feed
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.