All 35 models in the Guide, side by side. Click any column header to sort. ← Back to Guide
| Model | Tier | Total Params | Active | Context | License | GPQA | SWE-Bench | Terminal-Bench | MMLU | Key Capabilities |
|---|---|---|---|---|---|---|---|---|---|---|
| Kimi K3 | 🌌 Frontier | 2.8T | ~50B | 1,048,576 | MIT | 93.5% | 76.8% | — | — | Coding, Reasoning, Long Context, Multimodal |
| DeepSeek V4 Pro | 🌌 Frontier | 1.6T | 49B | 1,000,000 | MIT | 90.1% | 80.6% | — | — | Coding, Reasoning, Math, Long Context |
| GLM-5.2 | 🌌 Frontier | — | — | 1,000,000 | Open Weights | 91.2% | — | 81.0 | — | Coding, Reasoning, Speed, Terminal Use |
| GLM-5.3 | 🌌 Frontier | 753B | 40B | 1,000,000 | Custom | — | — | 28.3 | — | Coding, Cybersecurity, Agentic, Long Context |
| Llama 4 Maverick | 🌌 Frontier | 400B | 17B | 1,000,000 | Llama Community | — | — | — | 85.5% | General Chat, Coding, Vision, Long Context |
| Mistral Large 3 | 🌌 Frontier | 675B | 41B | 256,000 | Apache 2.0 | — | — | — | — | Multilingual, Coding, Reasoning, EU Compliant |
| Qwen3.8-2.4T-A95B | 🌌 Frontier | 2.4T | 95B | 262K → 1M | Custom | 92.6% | — | — | — | Coding, Reasoning, Long Context, Agentic |
| DeepSeek V4.1-Flash | 🌌 Frontier | 552B + 196B Engram | 8B prefill / 16B decode | 1,048,576 | MIT | 90.9% | — | 90.6 | — | Coding, Reasoning, Long Context, Agentic |
| Tencent Hy4 Preview | 🌌 Frontier | 770B | 49B | 1,000,000+ | Apache 2.0 | 92.3% | 65.7% | 85.4 | — | Coding, Reasoning, Long Context, Agentic |
| K2 Horizon (375B-A23B) | 🌌 Frontier | 375B | 23B | 524K | Apache 2.0 | 87.3% | — | 66.9 (audited) | — | Coding, Reasoning, Agentic, Fully Open |
| Atria Dawn Preview | 🌌 Frontier | 744B | ~40B | 256,000 | MIT | — | 59.6 | — | — | Agentic, Coding, Research, Tool Use |
| MiMo-V2.6-Pro | 🌌 Frontier | 1.02T | ~42B | 1,048,576 | MIT | — | 71.9 | — | — | Agentic, Coding, Multimodal, Long Context |
| Qwen 3.5 (397B-A17B) | 🚀 Mid-Tier | 397B | 17B | 128K-1M | Apache 2.0 | — | — | — | — | Multilingual, Vision, Long Context, Math |
| Qwen3-Coder (480B-A35B) | 🚀 Mid-Tier | 480B | 35B | 256K-1M | Apache 2.0 | — | 73.4% | — | — | Agentic Coding, Code Gen, Long Context, Tool Use |
| DeepSeek V4 Flash | 🚀 Mid-Tier | 284B | 13B | 1,000,000 | MIT | — | — | — | — | Coding, Reasoning, Long Context, Cost-Efficient |
| Mistral Medium 3.5 | 🚀 Mid-Tier | 128B | — | 256,000 | Modified MIT | — | — | — | — | Coding, Reasoning, Instruction Following, EU Compliant |
| Command A+ (218B-A25B) | tier-badge tier-mid">🚀 Mid-Tier | 218B | 25B | 131,072 | Apache 2.0 | 76.1%* | — | 25% (Hard)* | — | Agentic, Vision, Multilingual (48), Enterprise |
| MiMo-V2.6-Flash | 🚀 Mid-Tier | 309B | ~15B | 1,048,576 | MIT | — | 67.9 | — | — | Agentic, Coding, Multimodal, Cybersecurity |
| GLM-5.3-Flash | 🚀 Mid-Tier | 320B | ~18B | 1,048,576 | MIT | — | 63.4 | 84.3 | — | Agentic, Coding, Multimodal, Long Context |
| Qwen3.8-Flash-Next (180B-A6B) | tier-badge tier-mid">🚀 Mid-Tier | 180B | 6B | 262,144 | Qwen Community | 91.7%* | 62.5% (Pro)* | — | 90.4% (Base)* | Agentic, Vision, Long Context, Qwen4 Architecture |
| Llama 4 Scout | 🛸 Small | 109B | 17B | 10,000,000 | Llama Community | — | — | — | — | Long Context, Vision, General Chat, Single GPU |
| Mistral Small 4 | 🛸 Small | 119B | 6.5B | 256,000 | Apache 2.0 | — | — | — | — | Reasoning, Vision, Agentic Coding, Single GPU |
| Qwen3.8-27B | 🛸 Small | 27B (dense) | — | 262K → 1M | Apache 2.0 | 89.2% | 61.7% | — | — | Reasoning, Vision, Video, Coding |
| Muse Glimmer 30B | 🛸 Small | 29.6B (dense) | — | 131,072 (128K) | Apache 2.0 | — | 76.0% | — | — | Agentic, Coding, Tool Use, Vision |
| Gemma 4 31B | tier-badge tier-small">🛸 Small | 30.7B | — | 262,144 | Apache 2.0 | 84.3%* | — | — | 85.2% (Pro)* | Vision, Multilingual, Single GPU, QAT Checkpoints |
| Gemma 3 27B | 🛸 Small | 27B | — | 128,000 | Gemma Terms | — | — | — | ~75% | Vision, Multilingual, Single GPU, Consumer Hardware |
| DeepSeek R1 | 🛸 Small | 671B (full) / 32B (distill) | 37B (full) | 128,000 | MIT | — | — | — | — | Math, Reasoning, Chain-of-Thought, Logic |
| Granite 4.2 (30B) | 🛸 Small | 30B (dense) | — | 128K (512K ext.) | Apache 2.0 | 66.4% | 57.0% | — | — | Reasoning, Coding, Agentic RL, Tool Use |
| Phi-4 (14B) | 📡 Edge | 14B | — | 16,000 | MIT | — | — | — | 80.4% | Coding, Reasoning, CPU Capable, Laptop |
| Qwen 3.5-35B-A3B | 📡 Edge | 35B | 3B | 262,000 | Apache 2.0 | — | — | — | — | Coding, General Chat, Consumer GPU, Long Context |
| Kolibri (78B-A3.5B) | 🛸 Small | 78.1B | 3.46B | 262,144 | Apache 2.0 | 84.3%* | 66.4%* | — | — | German/English, Sovereign AI, Long Context, Abstention Trained |
| North Mini Code (30B-A3B) | tier-badge tier-edge">📡 Edge | 30B | 3B | 262,144 | Apache 2.0 | — | 67.6%* | — | — | Agentic Coding, Terminal, Long Context, Consumer GPU |
| K2 Horizon 7B | 📡 Edge | 7B (dense) | — | 32K+ | Apache 2.0 | — | 70.6% | — | — | Coding, Reasoning, Agentic, Consumer GPU |
| Nemotron 3 Nano (30B) | 📡 Edge | 30B | — | 1,000,000 | NVIDIA Open | 73.0% | — | — | 78.3% | Long Context, NVIDIA Optimised, Consumer GPU, Edge |
| MiniCPM5-2B | 📡 Edge | 2.52B (dense) | — | 131,072 | Apache 2.0 | — | 46.4% | — | — | On-Device, Coding, Tool Use, Reasoning |
| Ternary Bonsai 2 27B | 📡 Edge | 27.36B (dense) | — | 262,144 | Apache 2.0 | — | 60.8% | — | — | Tiny Footprint, Coding, Reasoning, Consumer GPU |
* vendor-reported scores (model-card or lab-blog numbers, high reasoning effort). Not yet independently verified.