MODEL MAKERS
Who makes the models.
Every maker with models on the site: how many it has, how many are tested, and the best score each reaches on the capability index.
Anthropic40 models, 22 tested167.3Claude Opus 5.5
OpenAI107 models, 53 tested166.4GPT-6 Astra
Moonshot AI17 models, 8 tested157.4Kimi K3
Google79 models, 41 tested157.3Gemini 3.7 Flash
Meta34 models, 11 tested156.8Muse Spark 1.3
xAI33 models, 16 tested156.4Grok 4.6
Alibaba130 models, 51 tested156.4Qwen 3.8 Max
Z.ai37 models, 14 tested155.6GLM-5.3
DeepSeek38 models, 24 tested155.3DeepSeek V4 Pro 0813
MiniMax17 models, 5 tested146.9MiniMax M3
NVIDIA24 models, 5 tested146.2Nemotron 3 Ultra
Mistral AI72 models, 24 tested141.3Mistral Medium 3.5
Microsoft8 models, 1 tested130.4Phi-4
Amazon10 models, 1 tested123.8Amazon Nova Pro
ByteDance36 models, 3 testedNot on the index yet
Cohere19 models, 4 testedNot on the index yet
Xiaomi14 models, 7 testedNot on the index yet
StepFun10 models, 3 testedNot on the index yet
Upstage7 models, 2 testedNot on the index yet
Baidu5 models, 1 testedNot on the index yet
Perplexity5 models, 0 testedNot on the index yet
Black Forest Labs4 models, 4 testedNot on the index yet
Inception4 models, 1 testedNot on the index yet
Meituan4 models, 1 testedNot on the index yet
Nous Research4 models, 0 testedNot on the index yet
Tencent3 models, 2 testedNot on the index yet
IBM2 models, 1 testedNot on the index yet
A maker is listed with at least three models that it sells, that several hosts list or that a test covers, or with one tested model. The score is its best model on the Epoch Capabilities Index, from Epoch AI (CC BY 4.0). Models from models.dev (MIT).