The AI Value Leaderboard: Intelligence per Dollar

Benchmarks tell you which model is smartest. Prices tell you which is cheapest. This table answers the question that matters: which model gives you the most capability for your money?

Value rankModelLabValue ScoreArena ScoreArena rankBlended $/1M
#1DeepSeek V4 Flash 0423openDeepSeek3314.21432 ±4#42$0.07
#2gpt-oss-120bOpenAI2355.91366 ±4#80$0.07
#3Qwen3 30B A3B Instruct 2507openQwen2181.81384 ±5#70$0.08
#4Gemma 3 12BGoogle1789.31334 ±9#92$0.08
#5Qwen3.5-FlashopenQwen1738.91398 ±4#64$0.11
#6Gemma 4 26B A4B Google1706.21435 ±8#41$0.14
#7gpt-oss-20bOpenAI1594.51288 ±6#102$0.06
#8Gemma 4 31BGoogle1510.61442 ±8#33$0.16
#9Gemma 3 4BGoogle1452.81291 ±9#101$0.06
#10Gemma 3n 4BGoogle14161306 ±5#99$0.08
#11Qwen3 32BopenQwen1077.71340 ±9#90$0.13
#12Gemini 2.5 Flash Lite Preview 09-2025Google1024.61379 ±3#71$0.18
#13Gemini 2.5 Flash LiteGoogle966.31369 ±4#77$0.18
#14Gemma 3 27BGoogle917.71358 ±4#84$0.17
#15GPT-5 NanoOpenAI874.91320 ±7#96$0.14
#16DeepSeek V3.2openDeepSeek774.51425 ±4#45$0.29
#17Qwen3 Next 80B A3B InstructopenQwen624.91419 ±5#49$0.35
#18Mistral Small 3openMistral584.31234 ±6#114$0.06
#19Qwen3 30B A3BopenQwen544.21317 ±5#98$0.22
#20Llama 4 ScoutopenMeta535.31280 ±5#107$0.15
#21GPT-5.6 LunaOpenAI507.31428 ±6#43$0.45
#22DeepSeek V4 Pro 0423openDeepSeek485.71451 ±4#23$0.52
#23GPT-4.1 NanoOpenAI484.61285 ±8#105$0.18
#24Llama 3.3 70B InstructopenMeta482.61275 ±3#109$0.16
#25DeepSeek V3.1 TerminusopenDeepSeek478.51417 ±10#53$0.45
#26Qwen3.7 PlusopenQwen459.11457 ±5#18$0.56
#27MiniMax M3openMiniMax448.41435 ±5#40$0.52
#28DeepSeek V3.2 ExpopenDeepSeek434.81333 ±5#93$0.31
#29Qwen3 Next 80B A3B ThinkingopenQwen406.11368 ±6#78$0.41
#30Qwen3.5-35B-A3BopenQwen391.41396 ±4#66$0.5
#31MiniMax M2.7openMiniMax390.51405 ±4#61$0.52
#32Qwen3.5-27BopenQwen387.91408 ±4#59$0.54
#33Gemini 3.1 Flash LiteGoogle381.91415 ±4#54$0.56
#34GPT-5.4 NanoOpenAI374.11373 ±4#74$0.46
#35MiniMax M2.1openMiniMax3641391 ±5#68$0.52
#36Qwen3 VL 235B A22B InstructopenQwen349.11421 ±6#48$0.63
#37MiniMax M2.5openMiniMax336.11359 ±4#83$0.47
#38GPT-4o-mini (2024-07-18)OpenAI329.91287 ±4#104$0.26
#39Qwen3 Coder 480B A35BopenQwen329.71357 ±5#85$0.48
#40Qwen-PlusopenQwen324.41327 ±8#95$0.39
#41Qwen3.6 PlusopenQwen323.81437 ±4#38$0.73
#42MiniMax M2openMiniMax318.71342 ±8#87$0.45
#43Qwen3.5-122B-A10BopenQwen304.81418 ±4#50$0.72
#44Qwen3 235B A22B Thinking 2507openQwen286.21414 ±6#56$0.75
#45Gemini 3.5 Flash LiteGoogle278.61437 ±6#39$0.85
#46Kimi K2.5openMoonshot AI272.81446 ±3#31$0.9
#47Qwen3.5 397B A17BopenQwen271.61438 ±4#36$0.88
#48Gemini 2.5 FlashGoogle255.81417 ±2#52$0.85
#49GPT-5 MiniOpenAI252.41374 ±5#73$0.69
#50Llama 4 MaverickopenMeta2501288 ±4#103$0.35
#51R1 0528openDeepSeek249.81428 ±6#44$0.91
#52Gemini 3 Flash PreviewGoogle236.41466 ±4#15$1.13
#53Qwen3 235B A22BopenQwen208.41366 ±5#79$0.8
#54GPT-4.1 MiniOpenAI200.91341 ±4#89$0.7
#55Kimi K2 ThinkingopenMoonshot AI199.31414 ±3#55$1.08
#56Mistral Small 3.1 24BopenMistral1941278 ±5#108$0.4
#57Phi 4Microsoft1921217 ±5#117$0.09
#58Nova 2 LiteAmazon190.71362 ±6#82$0.85
#59Gemini 3.6 FlashGoogle186.71480 ±6#7$1.50
#60Kimi K2 0711openMoonshot AI170.61371 ±5#76$1.00
#61Kimi K2 0905openMoonshot AI166.61379 ±7#72$1.08
#62Grok 4.20xAI160.41451 ±4#24$1.56
#63Grok 4.20 Multi-AgentxAI160.11450 ±4#26$1.56
#64Qwen3 VL 235B A22B ThinkingopenQwen154.51401 ±7#63$1.30
#65Qwen3 MaxopenQwen153.11439 ±4#35$1.56
#66Llama 3.1 70B InstructopenMeta152.51261 ±4#113$0.4
#67R1openDeepSeek150.21373 ±5#75$1.15
#68Kimi K2.6openMoonshot AI148.81455 ±5#19$1.71
#69MiniMax M1openMiniMax147.41342 ±4#88$0.96
#70Grok 4.3xAI126.41398 ±4#65$1.56
#71GPT-5.4 MiniOpenAI125.71412 ±4#57$1.69
#72Qwen3.7 MaxopenQwen1241474 ±10#13$2.21
#73Qwen3.6 Max PreviewopenQwen106.51446 ±8#29$2.31
#74Qwen3.8 MaxopenQwen97.21492 ±8#4$3.00
#75Claude Haiku 4.5Anthropic97.11394 ±3#67$2.00
#76Grok 4.5xAI83.61451 ±5#25$3.00
#77Gemini 3.5 FlashGoogle83.41482 ±5#6$3.38
#78Grok 4.6xAI82.11446 ±12#28$3.00
#79o4 MiniOpenAI79.21352 ±4#86$1.93
#80Gemini 2.5 ProGoogle74.91457 ±2#17$3.44
#81Mistral Medium 3.5openMistral73.61421 ±7#47$3.00
#82GPT-5.1OpenAI64.71422 ±4#46$3.44
#83GPT-5.6 SolOpenAI63.21453 ±6#20$4.00
#84Gemini 3.1 Pro PreviewGoogle62.21480 ±3#8$4.50
#85Gemini 3.1 Pro Preview Custom ToolsGoogle62.11480 ±3#9$4.50
#86Nano Banana Pro (Gemini 3 Pro Image)Google62.11479 ±4#11$4.50
#87Nano Banana Pro (Gemini 3 Pro Image Preview)Google62.11479 ±4#10$4.50
#88o3 MiniOpenAI621319 ±4#97$1.93
#89Claude Sonnet 5Anthropic60.81443 ±5#32$4.00
#90GPT-5OpenAI59.81406 ±4#60$3.44
#91GPT-5 ChatOpenAI59.21404 ±4#62$3.44
#92GPT-5.6 TerraOpenAI54.61446 ±6#30$4.50
#93GPT-5.2 ChatOpenAI49.71439 ±4#34$4.81
#94Gemma 2 27BGoogle48.51232 ±3#115$0.65
#95Kimi K3openMoonshot AI461476 ±6#12$6.00
#96GPT-5.4OpenAI44.81452 ±4#22$5.63
#97GPT-5.2OpenAI44.11412 ±3#58$4.81
#98Claude Sonnet 4.6Anthropic42.91458 ±4#16$6.00
#99Claude Sonnet 4.5Anthropic39.61438 ±3#37$6.00
#100GPT-5.3 ChatOpenAI391388 ±4#69$4.81
#101Claude Opus 5Anthropic30.51505 ±5#1$10.00
#102Command ACohere301331 ±3#94$4.38
#103Claude Opus 4.6Anthropic29.71497 ±3#2$10.00
#104Claude Opus 4.7Anthropic28.31483 ±4#5$10.00
#105Claude Opus 4.8Anthropic25.31453 ±5#21$10.00
#106Claude Opus 4.5Anthropic251450 ±3#27$10.00
#107GPT-5.5OpenAI23.71466 ±4#14$11.25
#108Claude Sonnet 4Anthropic231338 ±4#91$6.00
#109Mistral Large 2407openMistral22.11266 ±4#111$3.00
#110GPT-4o (2024-08-06)OpenAI18.91283 ±4#106$4.38
#111GPT-4.1OpenAI18.21264 ±4#112$3.50
#112Claude Fable 5Anthropic14.71493 ±5#3$20.00
#113GPT-4o (2024-05-13)OpenAI13.41301 ±3#100$7.50
#114Claude Opus 4.1Anthropic7.31418 ±3#51$30.00
#115Command R+ (08-2024)Cohere6.61229 ±7#116$4.38
#116Claude Opus 4Anthropic5.51364 ±4#81$30.00
#117GPT-4 TurboOpenAI4.81272 ±4#110$15.00

Methodology

Value Score = (Arena Score − 1200) ÷ blended price per 1M tokens, where blended price = 0.75 × input + 0.25 × output (a typical 3:1 token mix). The 1200-point baseline measures capability above a dated-model floor, so cheap-but-weak models can't win on price alone. A per-dollar metric still inherently favors efficient models over absolute frontier models — that's the point of this page; for raw capability, sort by Arena rank.

Capability data: Arena Scores © LMArena, licensed CC-BY-4.0, as of 12 Aug 2026. Prices: live market data, refreshed every 6 hours. No affiliate links on this page.