Skip to main content

BenchLM recommendation

Best Multimodal LLMs in 2026

Data verified

As of July 13, 2026, the top model in best multimodal llms on the BenchLM leaderboard is Gemini 3 Pro Deep Think with a score of 95.

Last verified: July 13, 2026

This page ranks models by BenchLM's multimodal & grounded category: image understanding, document and chart reading, and visually grounded reasoning. Note the scope: these scores measure how well a model understands images — not how well it generates them. Image-generation systems are a different product class and are not ranked here.

Unless noted otherwise, ranking surfaces on this page use BenchLM's provisional leaderboard lane rather than the stricter sourced-only verified leaderboard.

Bottom line: Gemini 3 Pro Deep Think leads multimodal understanding at 95, with Claude Mythos 5 and GPT-5.1 behind it. For image generation, look at dedicated image models — this table measures visual understanding.

According to BenchLM.ai, Gemini 3 Pro Deep Think leads this ranking with a score of 95, followed by Grok 4.1 (93.1) and Claude Mythos 5 (92.8). The top three are separated by just a few points — any of them would perform well for this use case.

The best open-weight option is GLM-5 (Reasoning) (ranked #27 with a score of 71.1). Proprietary models hold a clear advantage in this category, though open-weight options may suffice for less demanding use cases.

This ranking is based on provisional weighted averages across the scoring benchmarks in multimodalGrounded tracked by BenchLM.ai. For detailed model profiles, click any model name below. To compare two specific models head-to-head, use the "vs #" links.

What changed

Gemini 3 Pro Deep Think leads multimodal & grounded understanding at 95.

Claude Mythos 5 strong #2 at 92.8 on grounded visual reasoning.

GPT-5.1 still a top-3 multimodal row at 91.8 — and cheap at $1.25/$10.

How to choose

Full Rankings (110 models)

1
Gemini 3 Pro Deep Think
Google·Proprietary·2M

95

prov. avg

Grok 4.1
xAI·Proprietary·1M

93.1

tracked

3
Claude Mythos 5
Anthropic·Proprietary·1M+

92.8

prov. avg

4
GPT-5.1
OpenAI·Proprietary·200K

91.8

prov. avg

5
GPT-5.3 Codex
OpenAI·Proprietary·400K

91.4

prov. avg

6
Claude Sonnet 4.5
Anthropic·Proprietary·200K

90.5

prov. avg

7
GPT-5 (high)
OpenAI·Proprietary·128K

89.2

prov. avg

GPT-5 (medium)
OpenAI·Proprietary·128K

86.8

tracked

9
GPT-5.1-Codex-Max
OpenAI·Proprietary·400K

86.4

prov. avg

Grok 4.1 Fast
xAI·Proprietary·1M

86.1

tracked

11
GPT-5.2-Codex
OpenAI·Proprietary·400K

85.5

prov. avg

12
Claude Sonnet 4.6
Anthropic·Proprietary·200K

84.4

prov. avg

13
Gemini 3.1 Pro
Google·Proprietary·1M

82.8

prov. avg

14
Gemini 2.5 Pro
Google·Proprietary·1M

82.1

prov. avg

15
GPT-5.2
OpenAI·Proprietary·400K

80.8

prov. avg

16
Gemini 3 Pro
Google·Proprietary·2M

80.3

prov. avg

17
Claude Fable 5
Anthropic·Proprietary·1M+

79.7

prov. avg

18
Gemini 3.5 Flash
Google·Proprietary·1M

75.6

prov. avg

19
Claude Opus 4.6
Anthropic·Proprietary·1M

75.4

prov. avg

20
Claude 4.1 Opus
Anthropic·Proprietary·200K

75.2

prov. avg

Muse Spark
Meta·Proprietary·262K

74.6

tracked

22
Claude 4 Sonnet
Anthropic·Proprietary·200K

73.6

prov. avg

23
Qwen3.7 Plus
Alibaba·Proprietary·1M

72.8

prov. avg

24
Claude Haiku 4.5
Anthropic·Proprietary·200K

72

prov. avg

25
Grok 4
xAI·Proprietary·128K

71.4

prov. avg

26
Gemini 3 Flash
Google·Proprietary·1M

71.4

prov. avg

27
GLM-5 (Reasoning)
Z.AI·Open Weight·200K

71.1

prov. avg

28
Qwen3.6 Plus
Alibaba·Proprietary·1M

68.4

prov. avg

29
Kimi K2.6
Moonshot AI·Open Weight·256K

68

prov. avg

30
MiMo-V2-Flash
Xiaomi·Open Weight·256K

67.6

prov. avg

MiMo-V2.5
Xiaomi·Proprietary·1M

67.1

tracked

32
Kimi K2.5 (Reasoning)
Moonshot AI·Proprietary·128K

67

prov. avg

Mistral Large 3
Mistral·Proprietary·128K

66.8

tracked

o1-preview
OpenAI·Proprietary·200K

66.6

tracked

35
Claude Opus 4.8
Anthropic·Proprietary·1M

66.2

prov. avg

36
Claude 3.5 Sonnet
Anthropic·Proprietary·200K

66.1

prov. avg

37
o3-mini
OpenAI·Proprietary·200K

65

prov. avg

Claude 4.1 Opus Thinking
Anthropic·Proprietary·200K

65

tracked

39
Qwen3.5 397B
Alibaba·Open Weight·128K

64.9

prov. avg

Gemini 1.5 Pro
Google·Proprietary·2M

64.9

tracked

o3-pro
OpenAI·Proprietary·200K

64.1

tracked

42
GPT-4.1
OpenAI·Proprietary·1M

63.5

prov. avg

43
Qwen3.5-27B
Alibaba·Open Weight·262K

62.9

prov. avg

44
Qwen3.6-27B
Alibaba·Open Weight·262K

62.7

prov. avg

45
Qwen3.5-122B-A10B
Alibaba·Open Weight·262K

62.6

prov. avg

46
Kimi K2.5
Moonshot AI·Open Weight·256K

62.5

prov. avg

47
GPT-4o
OpenAI·Proprietary·128K

62

prov. avg

48
Qwen3.6-35B-A3B
Alibaba·Open Weight·262K

61.8

prov. avg

49
o3
OpenAI·Proprietary·200K

61.5

prov. avg

50
Qwen3.5-35B-A3B
Alibaba·Open Weight·262K

61.1

prov. avg

51
GPT-5.4
OpenAI·Proprietary·1.05M

60

prov. avg

52
Claude Opus 4.5
Anthropic·Proprietary·200K

59.3

prov. avg

53
DeepSeek V3.2 (Thinking)
DeepSeek·Open Weight·128K

59.2

prov. avg

54
o1
OpenAI·Proprietary·200K

59.1

prov. avg

Claude 3 Opus
Anthropic·Proprietary·200K

59.1

tracked

Qwen3.5 397B (Reasoning)
Alibaba·Open Weight·128K

58.6

tracked

57
GLM-4.7
Z.AI·Open Weight·200K

58.5

prov. avg

58
GPT-5.5
OpenAI·Proprietary·1M

58.3

prov. avg

59
Gemini 3.1 Flash-Lite
Google·Proprietary·1M

57.3

prov. avg

60
GPT-4.1 mini
OpenAI·Proprietary·1M

57.2

prov. avg

61
GLM-5
Z.AI·Open Weight·200K

56.6

prov. avg

Claude 3 Haiku
Anthropic·Proprietary·200K

56.4

tracked

Gemini 1.0 Pro
Google·Proprietary·32K

55.9

tracked

Qwen2.5-1M
Alibaba·Open Weight·1M

55.1

tracked

65
Gemini 2.5 Flash
Google·Proprietary·1M

54.8

prov. avg

DeepSeekMath V2
DeepSeek·Open Weight·128K

54.7

tracked

Qwen2.5-72B
Alibaba·Open Weight·128K

52.8

tracked

68
DeepSeek V3.2
DeepSeek·Open Weight·128K

51.3

prov. avg

69
Claude Opus 4.7 (Adaptive)
Anthropic·Proprietary·1M

49.7

prov. avg

70
MiniMax M3
MiniMax·Open Weight·1M

49.6

prov. avg

71
o4-mini (high)
OpenAI·Proprietary·200K

49.5

prov. avg

DeepSeek LLM 2.0
DeepSeek·Open Weight·128K

49.1

tracked

Llama 3.1 405B
Meta·Open Weight·128K

45.9

tracked

74
Grok 4.20
xAI·Proprietary·2M

45.6

prov. avg

GPT-4o mini
OpenAI·Proprietary·128K

43.6

tracked

Mistral Large 2
Mistral·Proprietary·128K

43.5

tracked

77
Nemotron 3 Super 100B
NVIDIA·Open Weight·1M

42.5

prov. avg

78
GPT-4.1 nano
OpenAI·Proprietary·1M

40.7

prov. avg

79
Llama 4 Scout
Meta·Open Weight·10M

39.4

prov. avg

DeepSeek Coder 2.0
DeepSeek·Open Weight·128K

39.2

tracked

81
Llama 4 Maverick
Meta·Open Weight·1M

37.8

prov. avg

Llama 4 Behemoth
Meta·Open Weight·32K

35.5

tracked

GPT-4 Turbo
OpenAI·Proprietary·128K

34.9

tracked

Command A+
Cohere·Open Weight·128K

33.9

tracked

Moonshot v1
Moonshot AI·Proprietary·128K

30.5

tracked

Llama 3 70B
Meta·Open Weight·128K

30.2

tracked

Z-1
Z·Proprietary·128K

27.1

tracked

Grok Code Fast 1
xAI·Proprietary·256K

26.1

tracked

Nemotron-4 15B
NVIDIA·Open Weight·32K

25.8

tracked

o1-pro
OpenAI·Proprietary·200K

24.4

tracked

91
GPT-OSS 120B
OpenAI·Open Weight·128K

24.1

prov. avg

Phi-4
Microsoft·Open Weight·16K

24.1

tracked

Mistral 8x7B
Mistral·Open Weight·32K

23.4

tracked

DeepSeek-R1
DeepSeek·Open Weight·128K

22.4

tracked

Nemotron 3 Nano 30B
NVIDIA·Open Weight·32K

18.5

tracked

Nemotron Ultra 253B
NVIDIA·Open Weight·32K

17.5

tracked

97
Grok 3 [Beta]
xAI·Proprietary·128K

15.8

prov. avg

98
Qwen3 235B 2507 (Reasoning)
Alibaba·Open Weight·128K

13.9

prov. avg

Gemma 3 27B
Google·Open Weight·32K

13.6

tracked

100
Qwen3 235B 2507
Alibaba·Open Weight·128K

13.3

prov. avg

DeepSeek V3.1 (Reasoning)
DeepSeek·Open Weight·128K

13

tracked

Nova Pro
Amazon·Proprietary·128K

12.4

tracked

GLM-4.5
Z.AI·Proprietary·128K

12.1

tracked

DeepSeek V3.1
DeepSeek·Open Weight·128K

10.8

tracked

GLM-4.5-Air
Z.AI·Proprietary·128K

10.1

tracked

106
GPT-OSS 20B
OpenAI·Open Weight·128K

8.8

prov. avg

DBRX Instruct
Databricks·Open Weight·32K

7.5

tracked

Mixtral 8x22B Instruct v0.1
Mistral·Open Weight·64K

7.5

tracked

Mistral 7B v0.3
Mistral·Open Weight·32K

7.5

tracked

Mistral 8x7B v0.2
Mistral·Open Weight·32K

7.5

tracked

These rankings update weekly

Get notified when models move. One email a week with what changed and why.

Free. No spam. Unsubscribe anytime.

Key Takeaways

The top model is Gemini 3 Pro Deep Think by Google with a provisional score of 95.

The best open-weight model is GLM-5 (Reasoning) at position #27.

110 models are included in this ranking.

Score in Context

What these scores mean

The multimodal & grounded score blends image understanding, document/chart reading, and visually grounded reasoning benchmarks. Higher means the model more reliably extracts and reasons over visual content.

Known limitations

These are understanding scores — they say nothing about image generation quality. OCR-heavy production workloads should also test resolution limits and page-count behavior, which benchmarks only partially cover.

Best Multimodal LLMs FAQ

What is the best multimodal LLM?

Gemini 3 Pro Deep Think currently leads BenchLM's multimodal & grounded category at 95, ahead of Claude Mythos 5 (92.8) and GPT-5.1 (91.8). The table above recomputes with every data refresh. For most teams GPT-5.1 is the value pick — top-3 understanding at $1.25/$10 per million tokens.

What is the best LLM for image generation?

This page ranks image understanding, not generation — LLM leaderboards measure how well models read and reason over images. Image generation is a separate product class (diffusion and autoregressive image models) that BenchLM's categories do not currently score. Treat any "LLM" ranking of image generation with skepticism.

Which LLM is best for reading documents and PDFs?

Document understanding tracks the multimodal & grounded category closely: Gemini 3 Pro Deep Think leads, and long-context support matters as much as vision quality for multi-hundred-page work — see the large-context rankings for models pairing 1M-token windows with strong vision.

Can these models understand charts and tables?

The top rows handle standard charts and tables well — chart-reading benchmarks feed this category. Reliability drops on dense dashboards, small text, and unusual chart types, so test your hardest real documents rather than trusting a single aggregate score.

Last updated: July 13, 2026

The AI models change fast. We track them for you.

A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.