Skip to content
Model Pareto

Writing & chat models ranked by quality, price and speed

BenchmarkInstruction followingCreative writingTranslationOther

Quality vs price

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

Lower price is better. Pareto frontier: Gemma 4 E4B, Qwen3.5 4B, Gemma 4 12B, Qwen3.5 9B, Gemma 4 26B A4B, GPT-6 Luna, Qwen3.8 Flash-Next, MiMo-V2.6-Pro, Step 5 Preview, GLM-5.3, Kimi K3, Claude Opus 5.5, Claude Opus 5, Claude Fable 5.1.

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

Ranking · 146 models

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

Writing & chat ranking. Column headers sort the table. Expand a row to see per-benchmark scores.
Select to compareSourceDetails
1
Claude Fable 5.1Frontier
97.6±3.5$20.0663/5 benchmarks measuredVerified
2
Claude Opus 5Frontier
93.4±4.3$10.0543/5 benchmarks measuredVerified
3
Claude Opus 5.5Frontier
93.0±4.3$8.00923/5 benchmarks measuredVerified
4
GPT-6 Astra
92.1±5.3$20.0523/5 benchmarks measuredVerified
5
Kimi K3FrontierOpen
89.1±4.3$6.00373/5 benchmarks measuredVerified
6
GLM-5.3FrontierOpen
86.0±4.3$2.15573/5 benchmarks measuredVerified
7
GPT-6 Sol
83.1±4.3$4.001103/5 benchmarks measuredVerified
8
Step 5 PreviewFrontierEstimated from other categories
82.6±12.7$1.43850/5 benchmarks measuredEstimated
9
Qwen3.8 Max (0902)
82.3±6.0$3.00392/5 benchmarks measuredVerified
10
Muse Spark 1.3
82.3±4.6$2.002193/5 benchmarks measuredVerified
11
MiMo-V2.6-ProFrontierOpen
81.8±6.0$0.54492/5 benchmarks measuredVerified
12
Qwen3.8 Flash-NextFrontierOpen
80.0±9.7$0.23541/5 benchmarks measuredLab-reported
13
Qwen3.8-Omni-Flash
78.8±9.7$0.23est.~541/5 benchmarks measuredLab-reported
14
GLM-5.3 FlashOpen
78.4±6.0$0.24462/5 benchmarks measuredVerified
15
Gemini 3.8 Flash
77.8±4.6$1.502913/5 benchmarks measuredVerified
16
GPT-5.5
77.0±3.3$11.3914/5 benchmarks measuredVerified
17
GPT-5.3 Codex
76.9±9.5$4.811451/5 benchmarks measuredVerified
18
Gemini 3.1 Pro
75.4±3.4$4.501205/5 benchmarks measuredMixed
19
DeepSeek V4 Pro (0813)Open
74.1±6.0$1.98972/5 benchmarks measuredVerified
20
Qwen3.8 2.4T A95BOpen
73.2±7.3$3.00382/5 benchmarks measuredMixed
21
Qwen3.7 Plus
72.9±4.2$0.70653/5 benchmarks measuredVerified
22
Grok 4.6
72.0±6.0$3.00602/5 benchmarks measuredVerified
23
GPT-5.6 Terra
71.7±3.3$4.50834/5 benchmarks measuredVerified
24
MiniMax-M3Open
71.0±4.2$0.521183/5 benchmarks measuredVerified
25
Grok 4.7
71.0±4.3$3.00403/5 benchmarks measuredVerified
26
DeepSeek V4.1 FlashOpen
69.8±4.3$0.522323/5 benchmarks measuredVerified
27
GPT-5.4 mini
69.1±4.2$1.692173/5 benchmarks measuredVerified
28
Qwen3.5 397B A17BOpen
69.0±4.2$1.35863/5 benchmarks measuredVerified
29
Claude Sonnet 5
68.3±4.3$4.00793/5 benchmarks measuredVerified
30
Hy3Open
67.7±6.0$0.25862/5 benchmarks measuredVerified
31
Intern-S2-Preview (35B-A3B)Open
67.7±9.7self-host~$0.046–0.38est.~1201/5 benchmarks measuredLab-reported
32
Xing4.0-29B-A4BOpen
67.4±9.7self-host~$0.067–0.60est.~1101/5 benchmarks measuredLab-reported
33
GPT-6 LunaFrontier
67.2±6.0$0.201322/5 benchmarks measuredVerified
34
Nova 2.0 Pro Preview
65.5±9.5$3.441201/5 benchmarks measuredVerified
35
Gemma 4 26B A4BFrontierOpen
64.8±4.5$0.16est.~893/5 benchmarks measuredVerified
36
o3
64.4±9.5$3.501431/5 benchmarks measuredVerified
37
Qwen3.8 27BOpen
64.3±3.6$1.13434/5 benchmarks measuredMixed
38
InklingOpen
63.8±3.6$1.761844/5 benchmarks measuredMixed
39
Gemma 4 31BOpen
63.5±3.6$0.21355/5 benchmarks measuredMixed
40
MiMo-V2.5Open
61.6±4.2$0.18403/5 benchmarks measuredVerified
41
Step 3.7 FlashOpen
61.6±9.5$0.441901/5 benchmarks measuredVerified
42
Muse GlimmerOpen
61.4±4.4$0.581334/5 benchmarks measuredMixed
43
Nova 2.0 Lite
61.2±9.5$0.851671/5 benchmarks measuredVerified
44
Gemini 3.5 Flash-Lite
61.0±4.3$0.853513/5 benchmarks measuredVerified
45
Nex-N2-ProOpen
60.7±9.5self-host~$0.20–1.76est.~921/5 benchmarks measuredVerified
46
Nemotron 3 UltraOpen
60.7±3.3$1.072154/5 benchmarks measuredVerified
47
Nemotron Cascade 2 30B A3BOpen
59.3±9.5self-host~$0.044–0.37est.~2101/5 benchmarks measuredVerified
48
Qwen3.5 122B A10BOpen
58.7±4.2$1.101293/5 benchmarks measuredVerified
49
Mistral Medium 3.5Open
58.3±4.2$3.001493/5 benchmarks measuredVerified
50
Qwen3.5 9BFrontierOpen
56.4±9.5$0.15601/5 benchmarks measuredVerified
51
Kimi K2.7 CodeOpen
56.4±9.5$1.71661/5 benchmarks measuredVerified
52
GPT-5.4 nano
55.6±4.2$0.461653/5 benchmarks measuredVerified
53
Nova 2.0 Omni
53.8±9.5$0.85est.~1701/5 benchmarks measuredVerified
54
KAT Coder Pro V2
53.2±9.5$0.52est.~1101/5 benchmarks measuredVerified
55
Apriel-v1.6-15B-ThinkerOpen
52.9±9.5self-host~$0.12–1.09est.~941/5 benchmarks measuredVerified
56
Qwen3.6 35B A3BOpen
52.0±9.5$0.841201/5 benchmarks measuredVerified
57
Command A+Open
51.0±9.5self-host~$0.19–1.562141/5 benchmarks measuredVerified
58
Nemotron 3 Super 120B A12BOpen
51.0±9.5$0.451641/5 benchmarks measuredVerified
59
EXAONE 4.5 33BOpen
50.5±9.5self-host~$0.17–1.49est.~841/5 benchmarks measuredVerified
60
HyperNova 60B 2605Open
50.4±9.5self-host~$0.066–0.59est.~1101/5 benchmarks measuredVerified
61
Qwen3.5 Omni Plus
50.3±9.5$1.50821/5 benchmarks measuredVerified
62
Claude Haiku 4.5
49.7±3.8$2.001094/5 benchmarks measuredMixed
63
DiffusionGemma 26B A4BOpen
47.2±9.5self-host~$0.062–0.56est.~891/5 benchmarks measuredVerified
64
K2 Think V2Open
45.3±9.5self-host~$0.36–3.22est.~741/5 benchmarks measuredVerified
65
Ling-2.6-1TOpen
45.1±9.5$0.85est.~871/5 benchmarks measuredVerified
66
Nemotron 3 Nano Omni 30B A3B ReasoningOpen
44.1±9.5$0.422751/5 benchmarks measuredVerified
67
K2-V2Open
44.0±9.5self-host~$0.35–3.09est.~751/5 benchmarks measuredVerified
68
Solar Open 100BOpen
42.9±9.5self-host~$0.12–0.95est.~971/5 benchmarks measuredVerified
69
Qwen3 Next 80B A3B InstructOpen
42.0±4.2$0.411583/5 benchmarks measuredVerified
70
Step3 VL 10BOpen
40.8±9.5self-host~$0.091–0.81est.~1201/5 benchmarks measuredVerified
71
Mistral Large 3Open
40.8±3.3$0.75774/5 benchmarks measuredVerified
72
Tri-21B-ThinkOpen
40.2±9.5self-host~$0.16–1.46est.~901/5 benchmarks measuredVerified
73
Falcon-H1R-7BOpen
39.9±9.5self-host~$0.074–0.66est.~1001/5 benchmarks measuredVerified
74
North Mini CodeOpen
39.5±9.5self-host~$0.044–0.37931/5 benchmarks measuredVerified
75
Trinity Large ThinkingOpen
39.4±4.2$0.393373/5 benchmarks measuredVerified
76
NVIDIA Nemotron 3 Nano 4BOpen
39.2±9.5self-host~$0.047–0.42est.~2001/5 benchmarks measuredVerified
77
gpt-oss-120bOpen
38.1±3.4$0.261865/5 benchmarks measuredMixed
78
Ring-2.6-1TOpen
38.1±9.5$0.851191/5 benchmarks measuredVerified
79
LFM2.5-8B-A1BOpen
37.4±9.5self-host~$0.031–0.26est.~1301/5 benchmarks measuredVerified
80
Mistral Small 4Open
37.2±9.5$0.261701/5 benchmarks measuredVerified
81
Magistral Small 1.2Open
36.9±9.5$0.75est.~1101/5 benchmarks measuredVerified
82
Gemma 4 12BFrontierOpen
36.6±9.5$0.151181/5 benchmarks measuredVerified
83
Cogito v2.1Open
35.4±9.5$1.25est.~821/5 benchmarks measuredVerified
84
LongCat Flash LiteOpen
32.6±9.5self-host~$0.059–0.52est.~1201/5 benchmarks measuredVerified
85
MiniCPM5-1BOpen
32.6±9.5self-host~$0.021–0.18est.~1401/5 benchmarks measuredVerified
86
Jamba Reasoning 3BOpen
32.4±9.5self-host~$0.041–0.37est.~1201/5 benchmarks measuredVerified
87
Qwen3.5 4BFrontierOpen
32.1±9.5$0.06231/5 benchmarks measuredVerified
88
NVIDIA Nemotron 3 Nano 30B A3BOpen
31.9±4.2$0.092453/5 benchmarks measuredVerified
89
gpt-oss-20bOpen
31.4±4.2$0.101813/5 benchmarks measuredVerified
90
Gemma 4 E4BFrontierOpen
30.2±9.5$0.04611/5 benchmarks measuredVerified
91
GLM-4.6VOpen
29.8±5.1$0.45683/5 benchmarks measuredVerified
92
LFM2 24B A2BOpen
29.5±9.5self-host~$0.051–0.42est.~1201/5 benchmarks measuredVerified
93
Nanbeige4.1-3BOpen
29.1±9.5self-host~$0.047–0.42est.~1101/5 benchmarks measuredVerified
94
HyperCLOVA X SEED Think (32B)Open
29.0±9.5self-host~$0.17–1.50est.~831/5 benchmarks measuredVerified
95
Llama 3.3 Instruct 70BOpen
28.5±9.5$0.71851/5 benchmarks measuredVerified
96
ERNIE 4.5 300B A47BOpen
28.1±9.5$0.49est.~791/5 benchmarks measuredVerified
97
Sarvam 105BOpen
28.0±9.5$0.07est.~991/5 benchmarks measuredVerified
98
INTELLECT-3Open
27.6±4.5self-host~$0.12–0.97est.~973/5 benchmarks measuredVerified
99
Llama 4 ScoutOpen
27.5±9.5$0.311071/5 benchmarks measuredVerified
100
Qwen3 Coder NextOpen
27.0±9.5$0.561261/5 benchmarks measuredVerified
101
Olmo 3 7B ThinkOpen
26.9±9.5self-host~$0.072–0.64est.~1001/5 benchmarks measuredVerified
102
LFM2.5-1.2B-InstructOpen
26.8±9.5self-host~$0.023–0.19est.~1401/5 benchmarks measuredVerified
103
Llama 3.1 Nemotron Ultra 253B v1Open
25.9±5.1self-host~$0.88–8.40est.~1103/5 benchmarks measuredVerified
104
LFM2.5-1.2B-ThinkingOpen
25.8±9.5self-host~$0.023–0.19est.~1401/5 benchmarks measuredVerified
105
Hermes 4 - Llama-3.1 405BOpen
25.6±9.5$1.50421/5 benchmarks measuredVerified
106
Devstral 2Open
25.6±9.5self-host~$0.53–5.07est.~841/5 benchmarks measuredVerified
107
Qwen3 Omni 30B A3B InstructOpen
25.5±9.5$0.43961/5 benchmarks measuredVerified
108
NVIDIA Nemotron Nano 12B v2 VLOpen
25.3±9.5$0.302011/5 benchmarks measuredVerified
109
Llama 4 MaverickOpen
24.7±3.3$0.421144/5 benchmarks measuredVerified
110
Ring-flash-2.0Open
23.9±4.5$0.25est.~1203/5 benchmarks measuredVerified
111
Command AOpen
23.6±9.5$4.38571/5 benchmarks measuredVerified
112
Olmo 3.1 32B InstructOpen
23.4±4.2self-host~$0.16–1.46est.~843/5 benchmarks measuredVerified
113
Olmo 3.1 32B ThinkOpen
23.3±4.2self-host~$0.16–1.46est.~843/5 benchmarks measuredVerified
114
Gemma 4 E2BOpen
22.8±9.5self-host~$0.035–0.32est.~961/5 benchmarks measuredVerified
115
Qwen3.5 2BOpen
22.8±9.5self-host~$0.030–0.27est.~951/5 benchmarks measuredVerified
116
EXAONE 4.0 32BOpen
22.6±9.5self-host~$0.16–1.46est.~841/5 benchmarks measuredVerified
117
Ministral 3 14BOpen
22.3±9.5$0.20871/5 benchmarks measuredVerified
118
Devstral Small 2Open
21.8±9.5self-host~$0.19–1.67est.~1101/5 benchmarks measuredVerified
119
Jamba 1.7 LargeOpen
21.0±9.5self-host~$0.55–4.92est.~711/5 benchmarks measuredVerified
120
Sarvam 30BOpen
20.2±9.5$0.05est.~1201/5 benchmarks measuredVerified
121
Olmo 3 7B InstructOpen
20.0±9.5$0.13est.~1001/5 benchmarks measuredVerified
122
Ministral 3 8BOpen
19.8±9.5$0.15871/5 benchmarks measuredVerified
123
MiniCPM-V 4.6 1.3BOpen
19.5±9.5self-host~$0.024–0.20est.~1401/5 benchmarks measuredVerified
124
Reka Flash 3Open
19.3±9.5$0.35est.~891/5 benchmarks measuredVerified
125
LFM2.5-VL-1.6BOpen
19.2±9.5self-host~$0.028–0.23est.~1301/5 benchmarks measuredVerified
126
Granite 4.0 H SmallOpen
19.1±9.5$0.11481/5 benchmarks measuredVerified
127
Hermes 4 - Llama-3.1 70BOpen
18.6±9.5self-host~$0.35–3.12est.~671/5 benchmarks measuredVerified
128
NVIDIA Nemotron Nano 9B V2Open
18.5±9.5$0.091551/5 benchmarks measuredVerified
129
Molmo2-8BOpen
18.5±5.1self-host~$0.081–0.72est.~1003/5 benchmarks measuredVerified
130
Jamba 1.7 MiniOpen
17.9±9.5self-host~$0.10–0.90est.~971/5 benchmarks measuredVerified
131
Llama 3.2 Instruct 11B (Vision)Open
17.7±9.5$0.34151/5 benchmarks measuredVerified
132
Nova Micro
17.3±9.5$0.062771/5 benchmarks measuredVerified
133
Ministral 3 3BOpen
16.9±9.5$0.102311/5 benchmarks measuredVerified
134
Kimi Linear 48B A3B InstructOpen
16.1±9.5self-host~$0.051–0.42est.~961/5 benchmarks measuredVerified
135
Exaone 4.0 1.2BOpen
16.1±9.5self-host~$0.024–0.20est.~1401/5 benchmarks measuredVerified
136
Granite 4.0 H 1BOpen
14.5±9.5self-host~$0.026–0.21est.~1201/5 benchmarks measuredVerified
137
Apertus 70B InstructOpen
14.3±9.5$1.34est.~751/5 benchmarks measuredVerified
138
Granite 4.0 MicroOpen
14.3±9.5self-host~$0.043–0.38est.~1101/5 benchmarks measuredVerified
139
Qwen3.5 0.8BOpen
13.9±9.5self-host~$0.018–0.15est.~1101/5 benchmarks measuredVerified
140
Apertus 8B InstructOpen
11.9±9.5$0.13est.~1001/5 benchmarks measuredVerified
141
Molmo 7B-DOpen
10.9±9.5self-host~$0.077–0.68est.~1001/5 benchmarks measuredVerified
142
Tiny Aya GlobalOpen
10.7±9.5self-host~$0.042–0.38est.~1301/5 benchmarks measuredVerified
143
Granite 4.0 H 350MOpen
9.5±9.5self-host~$0.011–0.088est.~1501/5 benchmarks measuredVerified
144
Granite 4.0 350MOpen
8.3±8.9self-host~$0.011–0.090est.~1501/5 benchmarks measuredVerified
145
Phi-4Open
5.8±4.2$0.22403/5 benchmarks measuredVerified
146
Gemma 3 270MOpen
5.0±7.3self-host~$0.009–0.077est.~1301/5 benchmarks measuredVerified

About writing & chat

Instruction following, creative writing, and translation. 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

Day-0 (model cards) Third-party

Top quality
Claude Fable 5.1
97.6 / 100
Cheapest on the frontier
Gemma 4 E4B
30.2 quality · $0.04
Newest
Claude Opus 5.5
Sep 22, 2026
Fully verified
92%
of 146 models