AI Compute RadarModels
Everytrackedmodel.
One page per model: what it needs in memory, where it runs, and how it moves on the radar — all from measured readings and the same fit engine as the finder.
49 of 49 shown
| # | Model | File | Popularity | Runs easily on | Runs at 8K |
|---|---|---|---|---|---|
| 1 | Gemma 4 E4BGoogle · general | 4.6 GB8B · Q4_K_M | more popular ▲ | a 12 GB card at 32K | 32 of 32 |
| 2 | Qwen3.8 Flash-NextAlibaba Qwen · general, coding, agents | 103.7 GB180B MoE · Q4_K | about the same ◆ | no tracked machine | 0 of 32 |
| 3 | Qwen3.8 27BAlibaba Qwen · general, coding, agents, research | 15.3 GB27.8B · Q4_K_M | about the same ◆ | a 32 GB card at 32K | 11 of 32 |
| 4 | gpt-oss 20BOpenAI · general, agents, research | 10.8 GB20.9B MoE · Q4_K_M | more popular ▲ | a 24 GB card at 32K | 19 of 32 |
| 5 | Llama 3.1 8BMeta · general, coding | 4.6 GB8B · Q4_K_M | more popular ▲ | unknown · attention shape unpublished | — |
| 6 | GLM-5.3-FlashZhipu AI · general, coding, agents | 186 GB321B · Q4_K | about the same ◆ | no tracked machine | 0 of 32 |
| 7 | GLM-5.3Zhipu AI · general, agents, research | 435.2 GB753B · Q4_K | about the same ◆ | no tracked machine | 0 of 32 |
| 8 | Gemma 4 31BGoogle · general, research | 17.1 GB31.3B · Q4_K_M | more popular ▲ | Mac mini M4 Pro 64GB at 32K | 7 of 32 |
| 9 | DeepSeek-R1DeepSeek · research | 376.7 GB685B MoE · Q4_K_M | about the same ◆ | no tracked machine | 0 of 32 |
| 10 | gpt-oss 120BOpenAI · general, research, agents | 58.5 GB117B MoE · Q4_K_M | more popular ▲ | NVIDIA DGX Spark at 32K | 1 of 32 |
| 11 | Ornith 1.5 35B-A3BOrnith AI · general, coding, agents | 20.2 GB36B MoE · Q4_K_M | more popular ▲ | Mac mini M4 Pro 64GB at 32K | 6 of 32 |
| 12 | Qwen3.8 2.4T-A95BAlibaba Qwen · general, agents, research | 1220.8 GB2.4T-A95B · IQ4_XS | more popular ▲ | no tracked machine | 0 of 32 |
| 13 | Gemma 4 12BGoogle · general, research | 6.6 GB12B · Q4_K_M | more popular ▲ | a 16 GB card at 32K | 27 of 32 |
| 14 | Ling 3.0 TinyInclusionAI · general, agents | 4.5 GB7.9B MoE · Q4_K_M | more popular ▲ | a 12 GB card at 32K | 32 of 32 |
| 15 | Qwen3 14BAlibaba Qwen · general, coding, research | 8.4 GB14.8B · Q4_K_M | about the same ◆ | a 32 GB card at 32K | 19 of 32 |
| 16 | Gemma 3 27BGoogle · general, research | 15.4 GB27.4B · Q4_K_M | more popular ▲ | Mac mini M4 Pro 64GB at 32K | 11 of 32 |
| 17 | Qwen3-Coder NextAlibaba Qwen · coding, agents | 45.1 GB79.7B MoE · Q4_K_M | more popular ▲ | NVIDIA DGX Spark at 32K | 3 of 32 |
| 18 | Qwen3 0.6BAlibaba Qwen · general | 0.6 GB0.6B · Q8_0 | fading ▼ | a 10 GB card at 32K | 32 of 32 |
| 19 | Qwen3-Coder 30B-A3BAlibaba Qwen · coding, agents | 17.3 GB30.5B MoE · Q4_K_M | about the same ◆ | Mac mini M4 Pro 64GB at 32K | 7 of 32 |
| 20 | Llama 3.3 70BMeta · general, research, agents | 39.6 GB70.6B · Q4_K_M | about the same ◆ | NVIDIA DGX Spark at 32K | 3 of 32 |
| 21 | DeepSeek-V4-ProDeepSeek · research, coding, agents | 791.3 GB1650B MoE · Q4_K | more popular ▲ | no tracked machine | 0 of 32 |
| 22 | Qwen3 8BAlibaba Qwen · general, coding, agents | 4.7 GB8.2B · Q4_K_M | fading ▼ | a 24 GB card at 32K | 28 of 32 |
| 23 | Hy3Tencent · general, research | 166.3 GB299B MoE · Q4_K_M | about the same ◆ | no tracked machine | 0 of 32 |
| 24 | Llama 3.2 3BMeta · general | 1.9 GB3B · Q4_K_M | fading ▼ | unknown · attention shape unpublished | — |
| 25 | Phi-4 MiniMicrosoft · general | 2.3 GB3.8B · Q4_K_M | fading ▼ | a 16 GB card at 32K | 32 of 32 |
| 26 | GLM-4.5-AirZhipu AI · coding, agents, general, research | 68 GB110B MoE · Q4_K_M | about the same ◆ | NVIDIA DGX Spark at 32K | 1 of 32 |
| 27 | Gemma 4 26B-A4BGoogle · general, research | 15.8 GB25.8B MoE · Q4_K_M | fading ▼ | a 32 GB card at 32K | 11 of 32 |
| 28 | LFM2.5 8B-A1BLiquid AI · general, agents | 4.8 GB8.5B MoE · Q4_K_M | more popular ▲ | a 12 GB card at 32K | 32 of 32 |
| 29 | KAT-Coder V2.5Kwaipilot · coding, agents | 19.9 GB34.7B MoE · Q4_K_M | more popular ▲ | Mac mini M4 Pro 64GB at 32K | 6 of 32 |
| 30 | Ornith 1.5 9BOrnith AI · general, coding | 5.4 GB9.7B · Q4_K_M | about the same ◆ | a 16 GB card at 32K | 28 of 32 |
| 31 | GLM-5.2Zhipu AI · coding, agents, research | 433.8 GB753B MoE · Q4_K_M | fading ▼ | no tracked machine | 0 of 32 |
| 32 | Granite 4.2 3BIBM · general, agents | 2.1 GB3.7B · Q4_K_M | more popular ▲ | a 11 GB card at 32K | 32 of 32 |
| 33 | DeepSeek-V4-FlashDeepSeek · general, coding, research | 144.4 GB304B MoE · Q4_K | about the same ◆ | no tracked machine | 0 of 32 |
| 34 | Qwen3 4BAlibaba Qwen · general, coding | 2.3 GB4B · Q4_K_M | fading ▼ | a 16 GB card at 32K | 32 of 32 |
| 35 | Qwen3 32BAlibaba Qwen · general, coding, research, agents | 18.4 GB32.8B · Q4_K_M | about the same ◆ | Mac mini M4 Pro 64GB at 32K | 6 of 32 |
| 36 | Ornith 1.5 397BOrnith AI · general, coding, research | 227.5 GB403B MoE · Q4_K_M | about the same ◆ | no tracked machine | 0 of 32 |
| 37 | GLM-4.7-FlashZhipu AI · coding, agents | 17.1 GB31.2B MoE · Q4_K_M | about the same ◆ | Mac mini M4 Pro 64GB at 32K | 11 of 32 |
| 38 | Devstral Small 2 24BMistral AI · coding, agents | 13.3 GB24B · Q4_K_M | about the same ◆ | Mac mini M4 Pro 64GB at 32K | 11 of 32 |
| 39 | Granite 4.2 8BIBM · general, coding, agents | 5 GB8.8B · Q4_K_M | fading ▼ | a 24 GB card at 32K | 28 of 32 |
| 40 | Granite 4.2 30BIBM · general, coding, agents | 16.5 GB29.3B · Q4_K_M | more popular ▲ | Mac mini M4 Pro 64GB at 32K | 7 of 32 |
| 41 | LFM2.5 1.2BLiquid AI · general | 0.7 GB1.2B · Q4_K_M | fading ▼ | a 8 GB card at 32K | 32 of 32 |
| 42 | Nemotron Nano 9B v2NVIDIA · general, research | 6.1 GB8.9B · Q4_K_M | more popular ▲ | a 32 GB card at 32K | 26 of 32 |
| 43 | Qwen AgentWorld 35B-A3BAlibaba Qwen · agents, coding | 20.6 GB34.7B MoE · Q4_K_M | fading ▼ | Mac mini M4 Pro 64GB at 32K | 6 of 32 |
| 44 | LFM2.5 2.6BLiquid AI · general, agents | 1.6 GB2.7B · Q4_K_M | fading ▼ | a 8 GB card at 32K | 32 of 32 |
| 45 | Ling 3.0 FlashInclusionAI · general, agents | 70.1 GB127B MoE · Q4_K_M | fading ▼ | NVIDIA DGX Spark at 32K | 1 of 32 |
| 46 | Mistral Small 3.2Mistral AI · general, agents | 13.3 GB24B · Q4_K_M | fading ▼ | Mac mini M4 Pro 64GB at 32K | 11 of 32 |
| 47 | LFM3 1.7BLiquid AI · general | 1 GB1.7B · Q4_K_M | collecting | a 8 GB card at 32K | 32 of 32 |
| 48 | Ornith 1.5 122BOrnith AI · general | 70.6 GB122B · Q4_K_M | collecting | NVIDIA DGX Spark at 32K | 1 of 32 |
| 49 | LFM2.5 24B-A2BLiquid AI · general, agents | 13.5 GB24B MoE · Q4_K_M | collecting | a 32 GB card at 32K | 11 of 32 |
Popularity is the 7-day direction of the score behind the board: downloads, trending and likes on Hugging Face, ranked against the other tracked models. Runs = easily or with ordinary headroom on that device at 8K context. Models without a measured open-weights file show no fit.