Local vision models
Image to text (VLM) · Understand images and video: describe, answer questions, read documents, detect and segment.
Quality · LMArena
Human-preference rating on image questions · higher is better
- Color = publisher
Source: LMArena · release of Oct 19, 2026 · checked Oct 25, 2026 · dataset · CC BY 4.0 · 143 groups have no rating in this release; they are left out, not scored as zero
GLM-5.3-Flash: 1,271 on LMArena (vision)
37 publishers in this task
Apache 2.0 (89)
134 of 162 groups, from Hub parameter counts
Choose your machine and the charts and tables mark what fits in its memory.
| GLM-5.3-Flash | 1,271 ± 6 | 321 | 240 | MIT | transformers | Sep 7, 2026 |
|---|---|---|---|---|---|---|
| gemma-4-31B-it | 1,262 ± 6 | 31.3 | 24 | Apache 2.0 | transformers · safetensors | Jul 20, 2026 |
| Qwen3.8-27B | 1,242 ± 8 | 27.8 | 21 | Apache 2.0 | transformers | Aug 14, 2026 |
| gemma-4-26B-A4B-it | 1,242 ± 6 | 25.8 | 19 | Apache 2.0 | transformers · safetensors | Jul 20, 2026 |
| Qwen3-VL-235B-A22B-Instruct | 1,215 ± 7 | 236 | 177 | Apache 2.0 | transformers · safetensors | Nov 26, 2025 |
| gemma-3-27b-it | 1,160 ± 8 | 27.4 | 21 | Publisher terms: Gemma | transformers · safetensors | Mar 21, 2025 |
| GLM-4.5V | 1,153 ± 12 | 108 | 81 | MIT | transformers · safetensors | Oct 25, 2025 |
| Llama-4-Maverick-17B-128E-Instruct | 1,145 ± 9 | 402 | 302 | Publisher terms | transformers · safetensors · pytorch | May 22, 2025 |
| step3 | 1,145 ± 12 | 321 | 241 | Apache 2.0 | transformers · safetensors | Jan 29, 2026 |
| Llama-4-Scout-17B-16E-Instruct | 1,128 ± 10 | 109 | 82 | Publisher terms | transformers · safetensors · pytorch | May 22, 2025 |
| Qwen2.5-VL-32B-Instruct | 1,119 ± 15 | 33.5 | 26 | Apache 2.0 | transformers · safetensors | Apr 14, 2025 |
| Molmo2-8B | 1,112 ± 20 | 8.7 | 7 | Apache 2.0 | transformers · safetensors | Jan 23, 2026 |
| Molmo-72B-0924 | 1,045 ± 13 | 73.3 | 55 | Apache 2.0 | transformers · safetensors | Oct 9, 2025 |
| Llama-3.2-90B-Vision-Instruct | 1,032 ± 9 | 88.6 | 67 | Publisher terms: Llama 3.2 | transformers · safetensors · pytorch | Mar 4, 2025 |
| Qwen2-VL-7B-Instruct | 1,031 ± 11 | 8.3 | 7 | Apache 2.0 | transformers · safetensors | Feb 6, 2025 |
| Molmo-7B-D-0924 | 994 ± 14 | 8 | 6 | Apache 2.0 | transformers · safetensors | Dec 15, 2025 |
| Llama-3.2-11B-Vision-Instruct | 993 ± 11 | 10.7 | 9 | Publisher terms: Llama 3.2 | transformers · safetensors · pytorch | Dec 4, 2024 |
| MiniCPM-V-2_6 | 962 ± 16 | 8.1 | 7 | Not declared | transformers · safetensors | Jun 13, 2025 |
| Phi-3.5-vision-instruct | 919 ± 16 | 4.1 | 4 | MIT | transformers · safetensors | Dec 10, 2025 |
| Molmo2-4B | Not on the LMArena vision leaderboard of 2026-10-19 | 4.9 | 4 | Apache 2.0 | transformers · safetensors | Jan 23, 2026 |
| MolmoMotion-4B-H1-F32 | Not on the LMArena vision leaderboard of 2026-10-19 | 4.9 | 4 | Apache 2.0 | transformers · safetensors | Jun 18, 2026 |
| MolmoMotion-4B-H3-F30 | Not on the LMArena vision leaderboard of 2026-10-19 | 4.9 | 4 | Apache 2.0 | transformers · safetensors | Jun 18, 2026 |
| olmOCR-2-7B-1025 | Not on the LMArena vision leaderboard of 2026-10-19 | 8.3 | 7 | Apache 2.0 | transformers · safetensors | Oct 22, 2025 |
| olmOCR-2-7B-1025-FP8 | Not on the LMArena vision leaderboard of 2026-10-19 | 8.3 | No estimate: a precision build of a listed model, packed weights, or a missing or partial Hub count | Apache 2.0 | transformers · safetensors | Feb 19, 2026 |