AI Model Comparison
Compare up to 5 AI models side-by-side with sticky columns, radar charts, benchmark overlays, capability matrix, and CSV/PDF export. Pricing, context windows, capabilities, and benchmarks all read from the centralized catalog.
Select Models to Compare
Capability Radar
Price per Mtok (Input)
How to Compare AI Models
Effective model comparison looks at five dimensions: price (input + output per million tokens), context window (how much input the model accepts), capabilities (vision, reasoning, tools, JSON, web search), benchmarks (MMLU, HumanEval, GPQA, MATH), and latency tier (flagship vs economy).
When to Pick a Flagship
Choose a flagship (GPT-5, Claude Opus 4, Gemini 2.5 Pro) for complex reasoning, long-context analysis, agentic workflows, and high-stakes content. Expect $2-15/Mtok input and $10-75/Mtok output.
When to Pick a Balanced Model
Choose a balanced model (GPT-4.1 Mini, Claude Sonnet 4, Gemini 2.5 Flash) for most production workloads: coding, chat, RAG, tool use. Expect $0.30-4/Mtok input and $1.60-15/Mtok output.
When to Pick an Economy Model
Choose an economy model (GPT-5 Nano, DeepSeek V3, Gemini 2.0 Flash Lite, Command R) for high-volume classification, routing, extraction, and simple chat. Expect $0.05-0.28/Mtok input and $0.30-0.60/Mtok output.
FAQ
How many models can I compare at once?
Up to 5 models. The radar chart overlays all selected models; the comparison table uses a sticky first column so you can scroll horizontally on mobile.
What benchmarks are shown?
MMLU (general knowledge), HumanEval (code generation), GPQA (graduate-level reasoning), and MATH (mathematical reasoning). Not all models have all benchmarks — missing values show as —.
Can I export the comparison?
Yes. Click Export CSV to download a spreadsheet or Export PDF to open a printable comparison table.
How is the capability score calculated?
The score is the percentage of 12 tracked capabilities (vision, reasoning, tools, parallel tools, JSON, streaming, caching, web search, audio in/out, PDF, multilingual) that the model supports.