Skip to content
Model Pareto

Vision models ranked by quality, price and speed

BenchmarkImage reasoningDocuments & charts

Quality vs price

10 models on the frontier — nothing else is both cheaper and better.

Lower price is better. Pareto frontier: Gemma 4 E4B, Qwen3.5 4B, Ling-3.0-flash-VL, MiMo-V2.5, GPT-6 Luna, Qwen3.8-Omni-Flash, Gemini 3.8 Flash, GPT-6 Sol, Claude Opus 5.5, GPT-6 Astra.

  • Best-value frontier (nothing is both cheaper and better)
  • Evidencestrong → weak
  • Estimated from other categories
  • Estimated (price or speed)

Ranking · 82 models

Price $/M tokens (blended 3:1) · Speed output tokens/s

Vision ranking. Column headers sort the table. Expand a row to see per-benchmark scores.
Select to compareSourceDetails
1
GPT-6 AstraFrontier
98.9±3.0$20.0523/6 benchmarks measuredVerified
2
Claude Opus 5.5Frontier
94.5±4.8$8.00923/6 benchmarks measuredVerified
3
Claude Fable 5.1
87.4±7.3$20.0662/6 benchmarks measuredVerified
4
GPT-6 SolFrontier
86.8±4.8$4.001103/6 benchmarks measuredVerified
5
Gemini 3.8 FlashFrontier
82.6±4.2$1.502914/6 benchmarks measuredMixed
6
Muse Spark 1.3
79.1±4.8$2.002193/6 benchmarks measuredVerified
7
Qwen3.8-Omni-FlashFrontier
78.6±10.2$0.23est.~541/6 benchmarks measuredLab-reported
8
GPT-5.6 Terra
78.1±4.8$4.50833/6 benchmarks measuredVerified
9
Claude Opus 5
77.4±4.8$10.0543/6 benchmarks measuredVerified
10
Qwen3.8 Max (0902)
77.3±4.8$3.00393/6 benchmarks measuredVerified
11
Kimi K3Open
77.0±3.9$6.00375/6 benchmarks measuredMixed
12
MiMo-V2.6-ProOpen
76.5±9.4$0.54491/6 benchmarks measuredVerified
13
Intern-S2-397BOpen
75.3±8.7self-host~$0.20–1.76est.~921/6 benchmarks measuredLab-reported
14
GPT-5.5
74.4±4.8$11.3913/6 benchmarks measuredVerified
15
GPT-5.3 Codex
73.3±8.2$4.811451/6 benchmarks measuredVerified
16
Qwen3.8 Flash-NextOpen
72.6±5.3$0.23543/6 benchmarks measuredMixed
17
Qwen3.8 27BOpen
69.0±5.3$1.13433/6 benchmarks measuredMixed
18
GPT-6 LunaFrontier
68.8±4.8$0.201323/6 benchmarks measuredVerified
19
Gemini 3.1 Pro
68.6±4.0$4.501204/6 benchmarks measuredVerified
20
Step 5 Preview
66.1±6.3$1.43852/6 benchmarks measuredVerified
21
Grok 4.6
65.2±7.3$3.00602/6 benchmarks measuredVerified
22
Intern-S2-Preview (35B-A3B)Open
64.3±8.7self-host~$0.046–0.38est.~1201/6 benchmarks measuredLab-reported
23
Grok 4.7
64.1±7.3$3.00402/6 benchmarks measuredVerified
24
MiniMax-M3Open
63.8±5.3$0.521183/6 benchmarks measuredMixed
25
Apodex 1.1
63.5±8.2$0.97est.~1001/6 benchmarks measuredVerified
26
Grok Build 0.1 0616
62.9±8.2$1.25551/6 benchmarks measuredVerified
27
Qwen3.7 Plus
62.9±6.3$0.70652/6 benchmarks measuredVerified
28
GLM-5.3 FlashOpen
62.1±7.3$0.24462/6 benchmarks measuredVerified
29
Step 3.7 FlashOpen
60.8±8.2$0.441901/6 benchmarks measuredVerified
30
Inkling SmallOpen
60.3±8.2$0.522061/6 benchmarks measuredVerified
31
MiMo-V2.5FrontierOpen
59.9±8.2$0.18401/6 benchmarks measuredVerified
32
Qwen3.5 397B A17BOpen
59.1±6.2$1.35862/6 benchmarks measuredVerified
33
GPT-5.4 mini
58.4±6.3$1.692172/6 benchmarks measuredVerified
34
Gemma 4 31BOpen
57.5±8.2$0.21351/6 benchmarks measuredVerified
35
DeepSeek V4.1 FlashOpen
56.4±4.8$0.522323/6 benchmarks measuredVerified
36
Claude Sonnet 5
56.2±4.2$4.00794/6 benchmarks measuredMixed
37
o3
56.0±8.2$3.501431/6 benchmarks measuredVerified
38
Ling-3.0-flash-VLFrontierOpen
54.2±6.3$0.111472/6 benchmarks measuredVerified
39
Qwen3.5 Omni Plus
53.3±8.2$1.50821/6 benchmarks measuredVerified
40
InklingOpen
52.3±4.2$1.761844/6 benchmarks measuredMixed
41
Gemma 4 26B A4BOpen
52.1±8.2$0.16est.~891/6 benchmarks measuredVerified
42
Gemini 3.5 Flash-Lite
51.3±4.2$0.853514/6 benchmarks measuredMixed
43
Qwen3.5 9BOpen
50.5±8.2$0.15601/6 benchmarks measuredVerified
44
Muse GlimmerOpen
49.0±3.9$0.581335/6 benchmarks measuredMixed
45
Qwen3.6 35B A3BOpen
48.6±6.3$0.841202/6 benchmarks measuredVerified
46
Nova 2.0 Pro Preview
48.5±8.2$3.441201/6 benchmarks measuredVerified
47
Nova 2.0 Lite
47.2±8.2$0.851671/6 benchmarks measuredVerified
48
EXAONE 4.5 33BOpen
47.1±8.2self-host~$0.17–1.49est.~841/6 benchmarks measuredVerified
49
Qwen3.5 122B A10BOpen
46.1±6.3$1.101292/6 benchmarks measuredVerified
50
GPT-5.4 nano
43.9±6.3$0.461652/6 benchmarks measuredVerified
51
DiffusionGemma 26B A4BOpen
43.5±8.2self-host~$0.062–0.56est.~891/6 benchmarks measuredVerified
52
Nova 2.0 Omni
42.2±8.2$0.85est.~1701/6 benchmarks measuredVerified
53
Step3 VL 10BOpen
39.2±8.2self-host~$0.091–0.81est.~1201/6 benchmarks measuredVerified
54
Mistral Medium 3.5Open
38.4±6.3$3.001492/6 benchmarks measuredVerified
55
Qwen3.5 4BFrontierOpen
37.5±8.2$0.06231/6 benchmarks measuredVerified
56
Gemma 4 12BOpen
36.5±8.2$0.151181/6 benchmarks measuredVerified
57
Claude Haiku 4.5
36.3±4.1$2.001094/6 benchmarks measuredMixed
58
Command A+Open
33.5±6.3self-host~$0.19–1.562142/6 benchmarks measuredVerified
59
Magistral Small 1.2Open
32.2±8.2$0.75est.~1101/6 benchmarks measuredVerified
60
Llama 4 MaverickOpen
32.1±6.7$0.421142/6 benchmarks measuredMixed
61
Qwen3 Omni 30B A3B InstructOpen
29.4±8.2$0.43961/6 benchmarks measuredVerified
62
Nemotron 3 Nano Omni 30B A3B ReasoningOpen
29.0±8.2$0.422751/6 benchmarks measuredVerified
63
Mistral Small 4Open
28.3±5.2$0.261703/6 benchmarks measuredVerified
64
Mistral Large 3Open
27.7±4.8$0.75773/6 benchmarks measuredVerified
65
NVIDIA Nemotron Nano 12B v2 VLOpen
27.2±8.2$0.302011/6 benchmarks measuredVerified
66
Llama 4 ScoutOpen
26.2±8.2$0.311071/6 benchmarks measuredVerified
67
Gemma 4 E4BFrontierOpen
25.9±8.2$0.04611/6 benchmarks measuredVerified
68
Ministral 3 14BOpen
23.5±6.3$0.20872/6 benchmarks measuredVerified
69
Ministral 3 8BOpen
21.1±6.2$0.15872/6 benchmarks measuredVerified
70
Devstral Small 2Open
20.6±6.3self-host~$0.19–1.67est.~1102/6 benchmarks measuredVerified
71
Qwen3.5 2BOpen
20.3±8.2self-host~$0.030–0.27est.~951/6 benchmarks measuredVerified
72
GLM-4.6VOpen
19.7±8.2$0.45681/6 benchmarks measuredVerified
73
Gemma 4 E2BOpen
19.6±8.2self-host~$0.035–0.32est.~961/6 benchmarks measuredVerified
74
Llama 3.2 Instruct 90B (Vision)Open
18.8±8.2self-host~$0.47–4.23est.~631/6 benchmarks measuredVerified
75
MiniCPM-V 4.6 1.3BOpen
17.6±8.2self-host~$0.024–0.20est.~1401/6 benchmarks measuredVerified
76
Molmo2-8BOpen
17.2±8.2self-host~$0.081–0.72est.~1001/6 benchmarks measuredVerified
77
Ministral 3 3BOpen
16.5±6.3$0.102312/6 benchmarks measuredVerified
78
Llama 3.2 Instruct 11B (Vision)Open
13.3±8.2$0.34151/6 benchmarks measuredVerified
79
LFM2.5-VL-1.6BOpen
11.8±8.2self-host~$0.028–0.23est.~1301/6 benchmarks measuredVerified
80
Qwen3.5 0.8BOpen
11.4±8.2self-host~$0.018–0.15est.~1101/6 benchmarks measuredVerified
81
Molmo 7B-DOpen
10.8±8.2self-host~$0.077–0.68est.~1001/6 benchmarks measuredVerified
82
Phi-4 Multimodal InstructOpen
5.9±7.1self-host~$0.060–0.53171/6 benchmarks measuredVerified

About vision

Image understanding, documents, and charts. Quality is a weighted average of the benchmarks below, each rescaled so 100 = the best tracked model. Benchmarks without a published score are estimated from related ones (shown dashed). Dot fill shows the evidence: solid = strong, tinted = medium, empty ring = weak (few benchmarks measured, lab-reported only or provisional). Methodology

Not comparable across labs (1)

Kept for reference, not counted in rankings: these use a lab-specific task set, answer key or scoring, so scores can't be compared fairly across labs. Only published scores are shown.

Day-0 (model cards) Third-party

Top quality
GPT-6 Astra
98.9 / 100
Cheapest on the frontier
Gemma 4 E4B
25.9 quality · $0.04
Newest
Claude Opus 5.5
Sep 22, 2026
Fully verified
83%
of 82 models