Free vision models

#	Model ↕	AA Index ↕	GPQA ↕	Context ↕	Size ↕	Free at ↕
1	Gemini 3.8 Flash Google VisionReasoningCodingTool use	40.9	95%	1M	Mid-size	Google AI Studio
2	Gemini 3.6 Flash Google VisionReasoningCodingTool use	34.0	93%	1M	Mid-size	UnoRouter
3	Gemini 3 Flash Google VisionReasoningCodingTool use	26.3	90%	1M	Mid-size	UnoRouter
4	Gemma 4 31B Google DeepMind · 31B VisionTool useOpen weights	19.0	86%	262K	Mid-size	OpenRouter, Together AI, NVIDIA Build, UnoRouter
5	Gemini 3.5 Flash Lite Google VisionReasoningCoding	22.2	84%	1M	Small	UnoRouter
6	Gemma 4 26B A4B Google DeepMind · 26B · 4B VisionTool useOpen weights	16.7	79%	262K	Mid-size	OpenRouter, Together AI, Cloudflare Workers AI, UnoRouter
7	Gemini 3.1 Flash Lite Google VisionReasoningCoding	15.6	82%	1M	Small	UnoRouter
8	Gemma 4 12B Google DeepMind · 12B VisionTool useOpen weights	14.2	75%	262K	Mid-size	Together AI, UnoRouter
9	Llama 4 Maverick Instruct Fp4 Meta · 400B · 17B VisionTool useOpen weights	—	—	1M	Frontier	Together AI
10	Nemotron 3 Nano Omni NVIDIA · 30B · 3B VisionReasoningTool useOpen weights	10.3	47%	256K	Mid-size	+1OpenRouter, Together AI, NVIDIA Build, TokenRouter, UnoRouter
11	Qwen3 VL 235B Alibaba Cloud · 235B · 22B VisionReasoningTool useOpen weights	13.4	77%	262K	Frontier	Together AI
12	Nemotron 3.5 Super VL NVIDIA VisionReasoningTool useOpen weights	—	—	1M	Mid-size	OpenRouter, UnoRouter, OpenCode Zen
13	Nemotron 3.5 Super VL 120B NVIDIA · 120B · 12B VisionReasoningTool useOpen weights	—	—	—	Frontier	NVIDIA Build
14	Qwen2.5 VL 72B Alibaba Cloud · 72B VisionTool useOpen weights	—	—	32K	Frontier	UnoRouter
15	Llama 4 Scout Meta · 109B · 17B VisionTool useOpen weights	8.1	59%	10.5M	Frontier	Together AI, Cloudflare Workers AI, UnoRouter
16	Claude Haiku 5.5 Anthropic VisionCodingTool use	—	—	1M	Mid-size	UnoRouter
17	Gemini 2.5 Flash Lite Google VisionReasoningCoding	8.5	63%	1M	Small	UnoRouter
18	Gemma 4 26B IT Chimerax Google DeepMind · 26B · 4B VisionTool useOpen weights	—	—	262K	Mid-size	UnoRouter
19	Gemma 4 26B IT Darksoul Google DeepMind · 26B · 4B VisionTool useOpen weights	—	—	262K	Mid-size	UnoRouter
20	Gemma 4 26B IT Luminous Google DeepMind · 26B · 4B VisionTool useOpen weights	—	—	262K	Mid-size	UnoRouter
21	Gemma 4 26B IT Meromero Google DeepMind · 26B · 4B VisionTool useOpen weights	—	—	262K	Mid-size	UnoRouter
22	Gemma 4 26B IT Moonlight Google DeepMind · 26B · 4B VisionTool useOpen weights	—	—	262K	Mid-size	UnoRouter
23	C4ai Aya Vision 32B Cohere · 32B VisionTool use	—	—	16K	Mid-size	Cohere
24	Claude Haiku 4.5 Anthropic · New VisionCodingTool use	—	—	200K	Mid-size	UnoRouter
25	Gemma 4 E4b Google DeepMind VisionOpen weights	8.9	58%	131K	Small	Together AI
26	Llama 3.2 11B Vision Meta · 11B VisionTool useOpen weights	5.4	22%	131K	Mid-size	Together AI, Cloudflare Workers AI, UnoRouter
27	GPT-4o OpenAI VisionTool use	8.4	54%	128K	Frontier	UnoRouter
28	Llama 3.2 90B Vision Meta · 90B VisionTool useOpen weights	6.4	43%	16K	Frontier	Together AI, NVIDIA Build
29	Gemma 4 E2b Google DeepMind VisionOpen weights	7.8	43%	131K	Small	Together AI
30	Gemma 3 4B Google DeepMind · 4B VisionOpen weights	4.8	29%	66K	Small	Together AI, NVIDIA Build
31	Command A Vision 07 2025 Cohere · 111B · Legacy Vision	—	—	128K	Frontier	Cohere
32	C clef-omni Vision	—	—	66K	Small	Cloudflare Workers AI
33	Cosmos Reason2 8B NVIDIA · 8B VisionReasoning	—	—	—	Small	NVIDIA Build
34	Qwen2.5 VL 7B Alibaba Cloud · 7B VisionOpen weights	—	—	33K	Small	UnoRouter
35	Gemma 3 27B Google DeepMind · 27B VisionTool useOpen weights	4.9	43%	66K	Mid-size	Together AI
36	Gemma 3 12B Google DeepMind · 12B VisionTool useOpen weights	3.8	35%	128K	Mid-size	NVIDIA Build
37	Neva 22B NVIDIA · 22B · Legacy Vision	—	—	—	Mid-size	NVIDIA Build
38	F Fuyu 8B Adept · 8B · Legacy Vision	—	—	—	Small	NVIDIA Build
39	K Kosmos 2 Microsoft · Legacy Vision	—	—	—	Small	NVIDIA Build
40	Phi 3 Vision 128K Microsoft · 4.2B · Legacy VisionOpen weights	—	—	131K	Small	NVIDIA Build
41	Vila NVIDIA · Legacy Vision	—	—	—	Small	NVIDIA Build

Benchmarks: Artificial Analysis (September 24, 2026). A dash means the model is not on their leaderboard; we never estimate scores ourselves.

FAQ

See also