Skip to main content
TokenCost logoTokenCost

The Best Value LLMs per Dollar

Which model buys the most benchmark quality per dollar?

18 models clear the 70-point floor and carry a metered price. The floor is an editorial line: without it the ratio always crowns the cheapest weak model. The cheapest of them outright is Nemotron 3 Ultra 550B at $1.00 per 1M tokens blended.

Method: Quality score divided by the blended price per 1M tokens (3:1 input-to-output tokens: 0.75 x input + 0.25 x output), among models scoring at least 70. Models priced at $0 (open weights with no metered API) are left out: you cannot divide by a price that does not exist. Pricing as of July 2026. 4 rows below carry a score measured at a non-default reasoning tier and are marked as such; every affected model is named on the hub. Read the full method.

RankModelQuality per $/1MQualityBlended $/1MEvidence
#1Nex-N2-ProNex AGI75.6075.6$1.00Both suites
#2Nemotron 3 Ultra 550BNVIDIA71.9071.9$1.00SWE-bench only
#3MiMo-V2-ProXiaomi52.0078.0$1.50SWE-bench only
#4Muse Spark 1.1Meta38.9577.9$2.00Terminal-Bench only
#5GLM-5.2Zhipu· non-default tier36.2377.9$2.15Terminal-Bench only
#6Grok 4.5xAI27.2081.6$3.00Terminal-Bench only
#7Gemini 3.6 FlashGoogle25.8377.5$3.00Terminal-Bench only
#8Gemini 3.5 FlashGoogle22.5876.2$3.38Terminal-Bench only
#9Claude Sonnet 5Anthropic· non-default tier20.8383.3$4.00Both suites
#10Qwen3.7 MaxAlibaba19.8774.5$3.75Terminal-Bench only
#11Kimi K3Moonshot14.1785.0$6.00Terminal-Bench only
#12GPT-5.6 TerraOpenAI12.8572.3$5.63Terminal-Bench only
#13Claude Sonnet 4.6Anthropic· non-default tier11.8771.2$6.00Terminal-Bench only
#14Claude Opus 4.8Anthropic8.7087.0$10.00Both suites
#15Claude Opus 4.7Anthropic· non-default tier8.3183.1$10.00Terminal-Bench only
#16GPT-5.6 SolOpenAI7.6586.1$11.25Terminal-Bench only
#17GPT-5.5OpenAI7.1680.5$11.25Terminal-Bench only
#18Claude Fable 5Anthropic4.2384.6$20.00Terminal-Bench only

Frequently asked

Nearby questions, answered elsewhere

All rankings