Best AI model for websites

In websites, work within 60 Elo of the best model is available for 4× less: MiMo-V2.6-Pro at $0.472 per thousand tasks against $1.98 for Muse Spark 1.3.

118 models rated head to head in this category, not scored by a general benchmark and assumed to transfer. Models are given a brief and build a complete page covering layout, copy, styling and responsiveness, then judged against each other by human preference.

Highest ratedMuse Spark 1.31361 Elo · $1.98 / 1k tasks
Best valueMiMo-V2.6-Pro1312 Elo · $0.472 / 1k tasks

Nothing cheaper is as good

7 models are undominated here: for each, no cheaper model scores within 18 Elo. Everything else in the table below is beaten on both counts at once.

  1. 1gpt-oss-20b857 Elo$0.066 / 1k131K contextfp4 on some hosts
  2. 2Ministral 3 3B 25121032 Elo$0.085 / 1k131K context
  3. 3Solar Mini 41179 Elo$0.088 / 1k524K context
  4. 4DeepSeek V4 Flash 04231211 Elo$0.170 / 1k1.05M contextfp4 on some hosts
  5. 5MiMo-V2.51270 Elo$0.183 / 1k1.05M context
  6. 6MiMo-V2.6-Pro1312 Elo$0.472 / 1k1.05M context
  7. 7Muse Spark 1.31361 Elo$1.98 / 1k1.05M context

Every model rated on websites

Elo measured in this category specifically. A model strong overall can sit well down this table, which is the whole reason for ranking by task.

#ModelElo$/1kContextHosts
1Muse Spark 1.31361$1.981.05M1
2Kimi K31339$6.241.05M25
3Claude Opus 5.51335$8.861.00M11
4Claude Sonnet 5.51327$4.031.00M8
5GPT-6.1 Sol1323$4.431.05M8
6Claude Fable 5.11317$19.981.00M4
7Muse Spark 1.21317$1.981.05M1
8Claude Opus 51313$11.251.00M11
9MiMo-V2.6-Pro1312$0.4721.05M4
10Gemini 3.7 Flash1308$1.541.05M6
11Gemini 3.8 Flash1305$1.541.05M6
12Gemini 3.6 Flash1303$1.541.05M7
13GLM 5.31302$1.931.05M41
14Claude Fable 51300$20.521.00M6
15Claude Opus 4.61295$10.261.00M6
16Grok 4.61293$3.40500K6
17GLM 5.21290$2.001.05M34
18GLM 5.11290$2.11203K13
19Claude Sonnet 4.61289$6.161.00M9
20Grok 4.51286$4.36500K4
21Claude Sonnet 51279$4.301.00M10
22Qwen3.7 Max1277$2.611.00M2
23GLM 5.3 Flash1276$0.2301.05M32
24Muse Spark 1.11273$1.981.05M1
25Qwen3.7 Plus1273$0.5841.00M2
26Kimi K2.61273$1.52262K17
27Claude Haiku 5.51272$0.2151.00M10
28MiMo-V2.5-Pro1271$0.4971.05M6
29Gemini 3.5 Flash1270$3.531.05M7
30MiMo-V2.51270$0.1831.05M5
31Kimi K2.7 Code1268$1.75262K12
32GLM 5 Turbo1267$1.95203K1
33MiniMax M31262$0.5471.05M13
34Claude Opus 4.81261$11.251.00M11
35GPT-5.51261$12.901.05M7
36Mistral Large 41260$1.001.05M3
37Grok 4.71258$3.40500K5
38Gemini 3.1 Pro Preview1257$4.701.05M6
39GLM 51252$1.42203K8
40Kimi K2.51252$1.22262K5
41Claude Opus 4.51251$10.26200K6
42Qwen3.6 Plus1246$0.9751.00M1
43MiniMax M2.71246$0.534197K6
44DeepSeek V4 Flash 07311242$0.2331.05M27
45DeepSeek V4 Pro 04231239$1.681.05M15
46GLM 5V Turbo1234$1.95203K1
47Grok 4.201234$2.172.00M4
48GLM 4.71229$0.980203K6
49Inkling1228$1.81524K2
50MiniMax M2.51227$0.523197K9
51GPT-5.41224$6.451.05M7
52DeepSeek V4 Flash 04231211$0.1701.05M16
53MiniMax M2.11206$0.706205K2
54Gemini 3 Flash Preview1201$1.181.05M6
55Grok 4.31200$2.171.00M4
56Step 3.7 Flash1199$0.470262K2
57GLM 4.7 Flash1199$0.193128K3
58GPT-5.21199$5.17400K4
59Qwen3.5 397B A17B1195$1.49262K10
60Claude Sonnet 4.51193$6.161.00M7
61Qwen3.5 Plus 2026-02-151191$0.7801.00M1
62GPT-5.11191$3.69400K5
63DeepSeek V3.1 Terminus1191$0.672164K2
64GPT-51189$3.69400K3
65Hy31185$0.265262K5
66DeepSeek V3.2 Exp1183$0.447164K1
67Solar Pro 41181$0.164524K2
68Claude Opus 4.11181$30.78200K1
69Solar Mini 41179$0.088524K2
70DeepSeek V3.21179$0.504164K12
71GLM 4.61179$0.954203K4
72GLM 4.51174$1.03131K1
73Gemini 2.5 Pro1170$3.691.05M7
74GPT-5.3-Codex1167$5.17400K3
75Mistral Large 3 25121167$0.760262K2
76GPT-5.1-Codex1166$3.69400K1
77Qwen3 Coder 480B A35B1163$0.666262K3
78R1 05281154$1.18164K4
79GLM 4.5 Air1151$0.343131K3
80Claude Sonnet 41150$6.16200K2
81MiniMax M21147$0.666205K2
82Nemotron 3 Ultra1145$1.09203K4
83Trinity Large Thinking1141$0.403262K1
84Mistral Medium 3.11137$0.821131K3
85GPT-5 Mini1129$0.738400K4
86Claude Haiku 4.51127$2.15200K8
87DeepSeek V3.11127$0.808164K5
88DeepSeek V31124$0.634128K2
89Kimi K2 Thinking1118$1.15262K2
90Gemini 2.5 Flash1118$0.9161.05M7
91GPT-5.1-Codex-Mini1115$0.742400K1
92Kimi K2 09051112$1.47262K1
93GPT-5 Nano1106$0.148400K4
94Qwen3 Coder 30B A3B Instruct1092$0.264262K2
95Gemini 3.1 Flash Lite Preview1087$0.5881.05M3
96Ministral 3 14B 25121086$0.170262K3
97Mistral Medium 31083$0.821131K3
98Ministral 3 8B 25121068$0.128262K3
99Qwen3 235B A22B Instruct 25071062$0.405262K7
100Qwen3 235B A22B Thinking 25071057$1.59128K1
101Kimi K2 07111055$1.37131K1
102GPT-4.11043$3.721.05M3
103o31041$3.72200K1
104Mercury 21034$0.363128K1
105Ministral 3 3B 25121032$0.085131K3
106Codestral 25081017$0.456256K2
107GPT-4.1 Mini1002$0.7441.05M3
108o4 Mini990$2.05200K1
109GPT-4.1 Nano977$0.1861.05M3
110gpt-oss-120b972$0.285131K24
111Qwen3 30B A3B959$0.29441K1
112Mistral Small 3.2 24B900$0.193128K4
113Llama 4 Maverick875$0.462524K4
114gpt-oss-20b857$0.066131K11
115Nova Premier 1.0840$5.401.00M1
116GPT-4o836$5.10128K2
117Nova Pro 1.0801$1.92300K2
118Llama 4 Scout755$0.393328K3

Other tasks

Elo comes from head-to-head build-offs via OpenRouter, compiled 12 October 2026 and refreshed daily. Cost is the median across every host serving each model, priced on 1,200 input and 300 output tokens with 60% cached. See the findings for method, or download the data.