Skip to main content

FindLLM Research

Reproducible market studies

Three compact views of the model market, calculated from the same governed catalog used by FindLLM comparisons. Each edition links to an immutable daily dataset so results can be checked and cited.

Price–quality frontier

Pareto-efficient models after the shared catalog gate, a Quality Index floor of 30, and a $0.01/M-token price floor. A model appears only when no cheaper candidate has equal or higher quality.

ModelQualityPrice/1M
DeepSeek: DeepSeek V4 FlashDeepSeek40.3$0.10
Nex AGI: Nex-N2-ProNex Agi41.0$0.44
MiniMax: MiniMax M3MiniMax44.4$0.52
Z.ai: GLM 5.2Z AI51.1$1.45
OpenAI: GPT-5.6 LunaOpenAI51.2$2.25
xAI: Grok 4.5xAI53.8$3.00
OpenAI: GPT-5.6 TerraOpenAI55.0$5.63
Anthropic: Claude Opus 4.8Anthropic55.7$10.00
OpenAI: GPT-5.6 SolOpenAI58.9$11.25
Anthropic: Claude Fable 5Anthropic59.9$20.00

Speed–quality frontier

Models that are not dominated on both published output speed and quality. A model appears only when no faster candidate has equal or higher quality.

ModelQualityOutput speed
StepFun: Step 3.7 FlashStepFun30.3404.4 tok/s
Google: Gemini 3.5 FlashGoogle50.2238.2 tok/s
OpenAI: GPT-5.6 LunaOpenAI51.2233.8 tok/s
OpenAI: GPT-5.6 TerraOpenAI55.0166.0 tok/s
OpenAI: GPT-5.6 SolOpenAI58.985.4 tok/s
Anthropic: Claude Fable 5Anthropic59.964.9 tok/s

Open-weight leaders

Highest-quality comparison-ready models classified as open and linked to a public model repository. Repository presence is not a substitute for reviewing the model license and use restrictions.

ModelQualityParameters / license
DeepSeek: DeepSeek V4 ProDeepSeek44.3mit
MoonshotAI: Kimi K2.6Kimi44.2other
MoonshotAI: Kimi K2.7 CodeKimi41.9other
DeepSeek: DeepSeek V4 FlashDeepSeek40.3mit
MoonshotAI: Kimi K2.5Kimi38.1other
Qwen: Qwen3.6 27BAlibaba37.127B
Qwen: Qwen3.5-27BAlibaba33.827B
Qwen: Qwen3.5 397B A17BAlibaba33.7397B
Qwen: Qwen3.5-122B-A10BAlibaba32.3122B
Qwen: Qwen3.6 35B A3BAlibaba31.635B
Google: Gemma 4 26B A4B Google30.326B