Compare

Put models head to head.

Pick up to four models and read them side by side — the best value in every row is called out. The selection lives in the URL, so any comparison is a link you can share.

Gemini 3.6 Flashnew×Gemini 3.5 Flash+Kimi K3new+Claude Fable 5new+Grok 4.5new+GPT-5.5×Gemini 3.5 Flash Litenew+Claude Opus 5new+GPT-5.6 Solnew+Sonar Reasoning Pro+DeepSeek V4 Flashnew+GPT-5.6 Lunanew+Claude Sonnet 5new+GPT-5.4 Mini+Grok 4.6new+GPT-5.6 Terranew+DeepSeek V3.2+Qwen3.7 Plus+Claude Opus 4.8+Grok 4.20+Mistral Medium 3.5new+Mistral Largenew+Sonar Pronew+GPT-5.4 Nano+Grok 4.3+DeepSeek V4 Pronew+DeepSeek R1+Llama 4 Mavericknew+Ministral 14B+Command Anew+Qwen3.8 Maxnew+Mistral Small+Nova 2 Litenew+Qwen3.7 Flashnew+GLM-5.1+Qwen3.7 Max+MiniMax M3new+Kimi K2.6new+Command R++GLM-5.2new+Nova Premiernew+Llama 4 Scoutnew+Llama 3.3 70B+2/4 selected

Metric	Gemini 3.6 Flash#1 · Gemini 3	GPT-5.5#6 · GPT-5.5
Accuracyhigher is better	78.9%	69.3%
Hallucinationlower is better	14.5%	14.8%
Cost / 1klower is better	$116	$303
Valueacc. pts per $/1k	0.68	0.23
p50 latencylower is better	52.8s	97.5s
Run-to-run σlower is better	±5.6	±2.1
Reliabilityhigher is better	100%	100%

Cost × accuracy

selected in focus