🤖 AI Model Comparison
Compare AI model pricing, context windows, and capabilities side-by-side — with a cost calculator for input and output tokens
Cost Calculator
| Model ▾ | Provider ▾ | Context ▾ | $ / 1M in ▾ | $ / 1M out ▾ | Type ▾ |
|---|
About this tool
When choosing an AI model for your application — whether for a chatbot, code generator, or data extraction pipeline — you need to balance price, context window, and capabilities. This tool compares 28+ models across 8 providers (OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Cohere) in a sortable, filterable table.
The cost calculator at the top lets you estimate real API costs: enter your expected input and output token counts and see the per-request price for each provider's flagship model. For example, a 1,000-token prompt with a 500-token response costs roughly $0.0063 with GPT-4o or $0.0090 with Claude Sonnet 4.
Filter by provider or model type (chat, reasoning, code, embedding, image) to narrow down to the models that fit your use case. Click column headers to sort — sort by $ / 1M out ascending to find the cheapest models for generation-heavy workloads.
FAQ
Why do reasoning models cost more than chat models? ›
Reasoning models (like o3, Claude Opus 4, DeepSeek-R1) use chain-of-thought — they generate internal reasoning tokens that are hidden from the user but still billed. A single reasoning call can produce thousands of hidden tokens, multiplying the effective cost. Chat models give you exactly the tokens you see. This is also why pricing tables usually show per-1M-token rates rather than per-request.
What does the context window number mean in practice? ›
The context window is the total amount of text (prompt + response) the model can process in one call. A 128K window fits roughly 300 pages of text. Models with 1M+ context (GPT-4.1, Gemini 2.5 Pro/Flash, Llama 4 Scout at 10M) can ingest entire codebases or book-length documents. But remember: longer context costs more — you pay for every input token, and attention computation scales quadratically.
Are the prices accurate? Should I check the official pages? ›
Prices listed are standard API rates as of July 2026 — but providers change pricing frequently, offer batch discounts (50% off for 24-hour turnaround), and have free tiers. Always verify at the provider's official pricing page before committing to production usage. Embedding and image models may use per-request or per-image pricing, which we approximate here.
What about open-source models — can I run them for free? ›
Models from Meta (Llama), Mistral, and DeepSeek have open weights you can self-host if you have the GPU hardware. The prices shown here are approximate hosted API costs from providers like Together AI, Groq, or Fireworks. Self-hosting eliminates per-token costs but requires infrastructure investment. For small to medium workloads, hosted APIs are usually cheaper when you factor in GPU time.