Skip to content
Model Pareto

Agents models ranked by quality, price and speed

BenchmarkTool useComputer useWeb research

Quality vs price

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

Lower price is better. Pareto frontier: Gemma 4 E4B, Sarvam 30B, Qwen3.5 4B, Sarvam 105B, Ling 3.0 Flash, MiMo-V2.6-Flash, Qwen3.8 Flash-Next, GLM-5.3 Flash, Muse Spark 1.3, Claude Opus 5.5.

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

Ranking · 173 models

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

Agents ranking. Column headers sort the table. Expand a row to see per-benchmark scores.
Select to compareSourceDetails
1
Claude Opus 5.5Frontier
99.6±4.1$8.00922/7 benchmarks measuredVerified
2
Claude Fable 5.1
89.6±5.8$20.0663/7 benchmarks measuredVerified
3
Muse Spark 1.3Frontier
87.4±5.8$2.002193/7 benchmarks measuredVerified
4
GLM-5.3Open
87.4±5.8$2.15573/7 benchmarks measuredVerified
5
Grok 4.7
86.7±7.8$3.00402/7 benchmarks measuredVerified
6
GLM-5.3 FlashFrontierOpen
85.5±5.1$0.24464/7 benchmarks measuredMixed
7
Qwen3.8 Max (0902)
85.2±5.8$3.00393/7 benchmarks measuredVerified
8
Claude Opus 5
84.8±4.6$10.0544/7 benchmarks measuredMixed
9
MiMo-V2.6-ProOpen
84.4±6.4$0.54493/7 benchmarks measuredMixed
10
Grok 4.6
84.1±5.8$3.00603/7 benchmarks measuredVerified
11
Qwen3.8 2.4T A95BOpen
83.0±5.8$3.00383/7 benchmarks measuredVerified
12
Kimi K3Open
82.4±3.5$6.00376/7 benchmarks measuredMixed
13
DeepSeek V4.1 FlashOpen
82.1±7.8$0.522322/7 benchmarks measuredVerified
14
Qwen3.8 Flash-NextFrontierOpen
81.7±5.8$0.23543/7 benchmarks measuredVerified
15
GPT-6 Astra
81.6±4.6$20.0524/7 benchmarks measuredMixed
16
Gemini 3.8 Flash
76.1±5.8$1.502913/7 benchmarks measuredVerified
17
GPT-6 Sol
76.0±7.8$4.001102/7 benchmarks measuredVerified
18
Step 5 Preview
75.4±6.3$1.43853/7 benchmarks measuredMixed
19
MiMo-V2.6-FlashFrontierOpen
74.8±10.7$0.18551/7 benchmarks measuredLab-reported
20
Qwen3.8 27BOpen
74.2±4.7$1.13434/7 benchmarks measuredMixed
21
Nex-N2.5-ProOpen
73.9±8.9self-host~$0.20–1.76est.~922/7 benchmarks measuredLab-reported
22
GPT-5.6 Terra
72.3±3.2$4.50836/7 benchmarks measuredMixed
23
DeepSeek V4 Pro (0813)Open
71.5±5.8$1.98973/7 benchmarks measuredVerified
24
Atria Dawn PreviewOpen
71.0±10.5self-host~$0.30–2.69est.~811/7 benchmarks measuredLab-reported
25
GPT-5.3 Codex
69.3±10.0$4.811451/7 benchmarks measuredVerified
26
GPT-5.5
67.4±3.2$11.3917/7 benchmarks measuredMixed
27
Claude Sonnet 5
65.3±3.8$4.00795/7 benchmarks measuredMixed
28
Motif 3Open
65.2±10.0self-host~$0.13–1.19est.~961/7 benchmarks measuredVerified
29
GPT-6 Luna
64.2±7.8$0.201322/7 benchmarks measuredVerified
30
K2 Horizon 375B A23BOpen
56.3±5.8self-host~$0.17–1.38est.~883/7 benchmarks measuredVerified
31
o3
54.9±10.0$3.501431/7 benchmarks measuredVerified
32
Gemini 3.1 Pro
54.7±3.4$4.501205/7 benchmarks measuredMixed
33
Qwen3.5 Omni Plus
53.5±10.0$1.50821/7 benchmarks measuredVerified
34
Ling 3.0 FlashFrontierOpen
49.8±10.0$0.113461/7 benchmarks measuredVerified
35
Nex-N2.5-MiniOpen
49.1±8.9self-host~$0.046–0.38est.~1202/7 benchmarks measuredLab-reported
36
Solar Pro 4
47.9±10.0$0.52931/7 benchmarks measuredVerified
37
Solar Open2 250BOpen
47.5±10.0self-host~$0.12–1.11est.~941/7 benchmarks measuredVerified
38
MiniMax-M3Open
47.4±3.4$0.521185/7 benchmarks measuredMixed
39
Nova 2.0 Pro Preview
47.2±7.8$3.441202/7 benchmarks measuredVerified
40
K2 Horizon MoVA 36B A4BOpen
47.2±5.8self-host~$0.053–0.44est.~1103/7 benchmarks measuredVerified
41
GPT-5.4 mini
47.2±4.2$1.692174/7 benchmarks measuredVerified
42
Ling-2.6-1TOpen
47.0±10.0$0.85est.~871/7 benchmarks measuredVerified
43
Apodex 1.1
46.6±5.8$0.97est.~1003/7 benchmarks measuredVerified
44
Nova 2.0 Omni
46.5±10.0$0.85est.~1701/7 benchmarks measuredVerified
45
Inkling SmallOpen
46.3±10.0$0.522061/7 benchmarks measuredVerified
46
Ling-3.0-flash-VLOpen
45.4±5.8$0.111473/7 benchmarks measuredVerified
47
Kimi K2.7 CodeOpen
45.3±4.2$1.71664/7 benchmarks measuredVerified
48
Hy3Open
45.1±4.6$0.25864/7 benchmarks measuredMixed
49
Quasar 438B
43.9±5.8$0.901523/7 benchmarks measuredVerified
50
G9v3-39A5BOpen
43.0±10.0self-host~$0.061–0.50est.~1101/7 benchmarks measuredVerified
51
Nova 2.0 Lite
42.6±7.8$0.851672/7 benchmarks measuredVerified
52
InklingOpen
42.5±4.6$1.761844/7 benchmarks measuredMixed
53
Qwen3.7 Plus
42.3±4.2$0.70654/7 benchmarks measuredVerified
54
Apriel-v1.6-15B-ThinkerOpen
41.7±10.0self-host~$0.12–1.09est.~941/7 benchmarks measuredVerified
55
HyperCLOVA X SEED Think (32B)Open
41.3±10.0self-host~$0.17–1.50est.~831/7 benchmarks measuredVerified
56
A.X-K2Open
41.1±10.0self-host~$0.25–2.21est.~841/7 benchmarks measuredVerified
57
EXAONE 4.5 33BOpen
41.1±7.8self-host~$0.17–1.49est.~842/7 benchmarks measuredVerified
58
Step 3.7 FlashOpen
41.0±5.8$0.441903/7 benchmarks measuredVerified
59
Gemini 3.5 Flash-Lite
40.6±4.7$0.853514/7 benchmarks measuredMixed
60
K2 Horizon 7BOpen
40.5±5.8self-host~$0.083–0.74est.~1003/7 benchmarks measuredVerified
61
Nex-N2-ProOpen
40.3±4.2self-host~$0.20–1.76est.~924/7 benchmarks measuredVerified
62
Tri-21B-ThinkOpen
39.9±10.0self-host~$0.16–1.46est.~901/7 benchmarks measuredVerified
63
GPT-5.4 nano
39.5±4.2$0.461654/7 benchmarks measuredVerified
64
MiniCPM-V 4.6 1.3BOpen
39.5±10.0self-host~$0.024–0.20est.~1401/7 benchmarks measuredVerified
65
GPT-5.5 Instant (June 2026)
39.3±5.8$11.31233/7 benchmarks measuredVerified
66
LongCat Flash LiteOpen
39.2±10.0self-host~$0.059–0.52est.~1201/7 benchmarks measuredVerified
67
Grok Build 0.1 0616
38.5±7.8$1.25552/7 benchmarks measuredVerified
68
MiniCPM5-1BOpen
38.0±10.0self-host~$0.021–0.18est.~1401/7 benchmarks measuredVerified
69
MiMo-V2.5Open
37.9±4.2$0.18404/7 benchmarks measuredVerified
70
Qwen3.5 2BOpen
37.7±10.0self-host~$0.030–0.27est.~951/7 benchmarks measuredVerified
71
Qwen3.5 397B A17BOpen
37.2±4.2$1.35864/7 benchmarks measuredVerified
72
Qwen3.5 122B A10BOpen
36.9±4.2$1.101294/7 benchmarks measuredVerified
73
Ring-2.6-1TOpen
36.0±4.2$0.851194/7 benchmarks measuredVerified
74
Ling 3.0 TinyOpen
35.9±10.0self-host~$0.029–0.24311/7 benchmarks measuredVerified
75
Muse GlimmerOpen
35.8±4.7$0.581334/7 benchmarks measuredMixed
76
Mistral Medium 3.5Open
35.0±3.4$3.001495/7 benchmarks measuredMixed
77
Qwen3.6 35B A3BOpen
34.3±4.2$0.841204/7 benchmarks measuredVerified
78
Qwen3.5 9BOpen
34.0±5.8$0.15603/7 benchmarks measuredVerified
79
Nemotron Cascade 2 30B A3BOpen
33.6±7.8self-host~$0.044–0.37est.~2102/7 benchmarks measuredVerified
80
KAT Coder Pro V2
32.8±5.8$0.52est.~1103/7 benchmarks measuredVerified
81
K-EXAONE 2.0 0803Open
32.3±10.0self-host~$0.29–2.58est.~821/7 benchmarks measuredVerified
82
K2 Horizon 3.7BOpen
32.2±5.8self-host~$0.056–0.50est.~1103/7 benchmarks measuredVerified
83
Solar Open 100BOpen
31.8±10.0self-host~$0.12–0.95est.~971/7 benchmarks measuredVerified
84
Qwen3.5 0.8BOpen
31.8±10.0self-host~$0.018–0.15est.~1101/7 benchmarks measuredVerified
85
Sarvam 105BFrontierOpen
31.4±10.0$0.07est.~991/7 benchmarks measuredVerified
86
Nemotron 3 UltraOpen
31.3±3.4$1.072155/7 benchmarks measuredMixed
87
Qwen3.5 4BFrontierOpen
30.8±7.8$0.06232/7 benchmarks measuredVerified
88
Gemma 4 31BOpen
30.4±4.2$0.21354/7 benchmarks measuredVerified
89
Trinity Large ThinkingOpen
29.5±4.2$0.393374/7 benchmarks measuredVerified
90
Gemma 4 26B A4BOpen
28.6±5.8$0.16est.~893/7 benchmarks measuredVerified
91
gpt-oss-120bOpen
28.3±4.2$0.261864/7 benchmarks measuredVerified
92
Nemotron 3 Super 120B A12BOpen
28.2±4.2$0.451644/7 benchmarks measuredVerified
93
Command A+Open
28.0±4.2self-host~$0.19–1.562144/7 benchmarks measuredVerified
94
LongCat 2.0Open
28.0±5.8$0.52est.~793/7 benchmarks measuredVerified
95
Nemotron 3 Nano Omni 30B A3B ReasoningOpen
28.0±7.8$0.422752/7 benchmarks measuredVerified
96
MiniCPM5-2BOpen
27.3±5.8self-host~$0.035–0.31est.~1203/7 benchmarks measuredVerified
97
HyperNova 60B 2605Open
27.0±5.8self-host~$0.066–0.59est.~1103/7 benchmarks measuredVerified
98
Qwen3 Coder NextOpen
26.9±4.2$0.561264/7 benchmarks measuredVerified
99
Gemma 4 12BOpen
26.7±10.0$0.151181/7 benchmarks measuredVerified
100
Nanbeige4.1-3BOpen
26.4±10.0self-host~$0.047–0.42est.~1101/7 benchmarks measuredVerified
101
K2-V2Open
26.1±10.0self-host~$0.35–3.09est.~751/7 benchmarks measuredVerified
102
Step3 VL 10BOpen
25.9±10.0self-host~$0.091–0.81est.~1201/7 benchmarks measuredVerified
103
INTELLECT-3Open
25.6±10.0self-host~$0.12–0.97est.~971/7 benchmarks measuredVerified
104
Falcon-H1R-7BOpen
25.4±10.0self-host~$0.074–0.66est.~1001/7 benchmarks measuredVerified
105
Granite 4.2 30BOpen
25.4±5.8$0.28773/7 benchmarks measuredVerified
106
K2 Think V2Open
25.1±7.8self-host~$0.36–3.22est.~742/7 benchmarks measuredVerified
107
Qwen3 Next 80B A3B InstructOpen
24.5±10.0$0.411581/7 benchmarks measuredVerified
108
gpt-oss-20bOpen
24.0±4.2$0.101814/7 benchmarks measuredVerified
109
Sarvam 30BFrontierOpen
23.8±10.0$0.05est.~1201/7 benchmarks measuredVerified
110
Hermes 4 - Llama-3.1 405BOpen
23.7±10.0$1.50421/7 benchmarks measuredVerified
111
GLM-4.6VOpen
22.4±10.0$0.45681/7 benchmarks measuredVerified
112
Claude Haiku 4.5
22.3±3.5$2.001095/7 benchmarks measuredMixed
113
Mercury 2.5
21.8±7.8$0.347812/7 benchmarks measuredVerified
114
Gemma 4 E4BFrontierOpen
21.7±10.0$0.04611/7 benchmarks measuredVerified
115
NVIDIA Nemotron 3 Nano 4BOpen
21.4±10.0self-host~$0.047–0.42est.~2001/7 benchmarks measuredVerified
116
Nemotron 3.5 LightningOpen
21.0±5.8$0.112673/7 benchmarks measuredVerified
117
NVIDIA Nemotron Nano 12B v2 VLOpen
20.8±10.0$0.302011/7 benchmarks measuredVerified
118
Qwen3 Omni 30B A3B InstructOpen
20.1±10.0$0.43961/7 benchmarks measuredVerified
119
Llama 3.3 Instruct 70BOpen
20.0±10.0$0.71851/7 benchmarks measuredVerified
120
Granite 4.2 8BOpen
19.8±5.8$0.11833/7 benchmarks measuredVerified
121
Llama 3.1 Nemotron Ultra 253B v1Open
19.5±10.0self-host~$0.88–8.40est.~1101/7 benchmarks measuredVerified
122
NVIDIA Nemotron Nano 9B V2Open
19.3±10.0$0.091551/7 benchmarks measuredVerified
123
Gemma 4 E2BOpen
19.3±10.0self-host~$0.035–0.32est.~961/7 benchmarks measuredVerified
124
NVIDIA Nemotron 3 Nano 30B A3BOpen
19.1±5.8$0.092453/7 benchmarks measuredVerified
125
North Mini CodeOpen
19.1±4.2self-host~$0.044–0.37934/7 benchmarks measuredVerified
126
Devstral 2Open
18.9±4.2self-host~$0.53–5.07est.~844/7 benchmarks measuredVerified
127
LFM2.5-8B-A1BOpen
18.9±10.0self-host~$0.031–0.26est.~1301/7 benchmarks measuredVerified
128
Olmo 3.1 32B InstructOpen
18.8±10.0self-host~$0.16–1.46est.~841/7 benchmarks measuredVerified
129
LFM2.5-2.6BOpen
18.8±10.0self-host~$0.037–0.33est.~1201/7 benchmarks measuredVerified
130
Devstral Small 2Open
18.6±4.2self-host~$0.19–1.67est.~1104/7 benchmarks measuredVerified
131
Hermes 4 - Llama-3.1 70BOpen
18.5±10.0self-host~$0.35–3.12est.~671/7 benchmarks measuredVerified
132
LFM2.5-1.2B-ThinkingOpen
18.3±10.0self-host~$0.023–0.19est.~1401/7 benchmarks measuredVerified
133
Exaone 4.0 1.2BOpen
18.3±10.0self-host~$0.024–0.20est.~1401/7 benchmarks measuredVerified
134
Granite 4.0 H 1BOpen
18.0±10.0self-host~$0.026–0.21est.~1201/7 benchmarks measuredVerified
135
Jamba Reasoning 3BOpen
17.9±10.0self-host~$0.041–0.37est.~1201/7 benchmarks measuredVerified
136
ERNIE 4.5 300B A47BOpen
17.8±10.0$0.49est.~791/7 benchmarks measuredVerified
137
Granite 4.0 H SmallOpen
17.6±10.0$0.11481/7 benchmarks measuredVerified
138
Command AOpen
17.3±10.0$4.38571/7 benchmarks measuredVerified
139
Ring-flash-2.0Open
17.1±10.0$0.25est.~1201/7 benchmarks measuredVerified
140
Granite 4.2 3BOpen
17.0±5.8$0.052203/7 benchmarks measuredVerified
141
Llama 3.2 Instruct 11B (Vision)Open
17.0±10.0$0.34151/7 benchmarks measuredVerified
142
Granite 4.0 H 350MOpen
16.9±10.0self-host~$0.011–0.088est.~1501/7 benchmarks measuredVerified
143
Jamba 1.7 LargeOpen
16.7±10.0self-host~$0.55–4.92est.~711/7 benchmarks measuredVerified
144
Nova Micro
16.7±10.0$0.062771/7 benchmarks measuredVerified
145
Ministral 3 14BOpen
16.6±4.2$0.20874/7 benchmarks measuredVerified
146
Granite 4.0 350MOpen
16.5±10.0self-host~$0.011–0.090est.~1501/7 benchmarks measuredVerified
147
Olmo 3 7B InstructOpen
16.5±10.0$0.13est.~1001/7 benchmarks measuredVerified
148
Apertus 70B InstructOpen
16.5±10.0$1.34est.~751/7 benchmarks measuredVerified
149
Jamba 1.7 MiniOpen
16.4±10.0self-host~$0.10–0.90est.~971/7 benchmarks measuredVerified
150
LFM2 24B A2BOpen
16.4±10.0self-host~$0.051–0.42est.~1201/7 benchmarks measuredVerified
151
Mistral Large 3Open
16.4±4.2$0.75774/7 benchmarks measuredVerified
152
Granite 4.0 MicroOpen
16.4±10.0self-host~$0.043–0.38est.~1101/7 benchmarks measuredVerified
153
LFM2.5-1.2B-InstructOpen
16.3±10.0self-host~$0.023–0.19est.~1401/7 benchmarks measuredVerified
154
Celeris-1
16.1±5.8$0.3315213/7 benchmarks measuredVerified
155
Apertus 8B InstructOpen
16.1±10.0$0.13est.~1001/7 benchmarks measuredVerified
156
EXAONE 4.0 32BOpen
15.9±10.0self-host~$0.16–1.46est.~841/7 benchmarks measuredVerified
157
Magistral Small 1.2Open
15.9±5.8$0.75est.~1103/7 benchmarks measuredVerified
158
Gemma 3 270MOpen
15.6±10.0self-host~$0.009–0.077est.~1301/7 benchmarks measuredVerified
159
LFM2.5-VL-1.6BOpen
15.5±10.0self-host~$0.028–0.23est.~1301/7 benchmarks measuredVerified
160
K2 Horizon 0.9BOpen
15.2±7.8self-host~$0.021–0.18est.~1402/7 benchmarks measuredVerified
161
Ministral 3 3BOpen
15.1±4.2$0.102314/7 benchmarks measuredVerified
162
Ministral 3 8BOpen
14.9±4.2$0.15874/7 benchmarks measuredVerified
163
Olmo 3.1 32B ThinkOpen
14.7±10.0self-host~$0.16–1.46est.~841/7 benchmarks measuredVerified
164
Mistral Small 4Open
14.6±3.4$0.261705/7 benchmarks measuredMixed
165
Olmo 3 7B ThinkOpen
14.3±10.0self-host~$0.072–0.64est.~1001/7 benchmarks measuredVerified
166
Phi-4Open
13.9±10.0$0.22401/7 benchmarks measuredVerified
167
Molmo2-8BOpen
13.9±10.0self-host~$0.081–0.72est.~1001/7 benchmarks measuredVerified
168
Reka Flash 3Open
13.7±10.0$0.35est.~891/7 benchmarks measuredVerified
169
Molmo 7B-DOpen
13.5±10.0self-host~$0.077–0.68est.~1001/7 benchmarks measuredVerified
170
Kimi Linear 48B A3B InstructOpen
13.5±10.0self-host~$0.051–0.42est.~961/7 benchmarks measuredVerified
170
Tiny Aya GlobalOpen
13.5±10.0self-host~$0.042–0.38est.~1301/7 benchmarks measuredVerified
172
Llama 4 MaverickOpen
13.3±5.8$0.421143/7 benchmarks measuredVerified
173
Llama 4 ScoutOpen
12.1±5.8$0.311073/7 benchmarks measuredVerified

About agents

Tool use, computer use, and web research. 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 (4)

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
Claude Opus 5.5
99.6 / 100
Cheapest on the frontier
Gemma 4 E4B
21.7 quality · $0.04
Newest
Claude Opus 5.5
Sep 22, 2026
Fully verified
86%
of 173 models