Which LLM should
you actually use?

Quality Score (0–100)

Based on Arena.ai ELO ratings (text arena). Min-max normalized: best model = 100, worst = 0.

Hover over any model's score to see the raw ELO and full calculation.

Data sources: text, code, and vision arenas.

Input Price per 1M tokens

What you pay to send tokens to the model (your prompt, system message, context).

1M tokens ≈ 750K words ≈ ~3 novels.

Example: sending a 4K-token prompt to a model priced at $3.00/1M costs $0.012

Output Price per 1M tokens

What you pay for tokens the model generates (its response). Usually 3–5× more expensive than input.

Example: a 1K-token response from a model priced at $15.00/1M costs $0.015

Blended Price per 1M tokens

Weighted average: 75% × Input + 25% × Output. Reflects a typical API usage pattern (prompts are usually longer than responses).

Example: Input $3.00, Output $15.00 → Blended = $3×0.75 + $15×0.25 = $6.00

Use this to compare total cost across models.

Value Score

Formula: Quality² ÷ Blended Price

Quality is squared so high-quality models are rewarded disproportionately. A model with score 90 at $3 beats score 60 at $1.

Example: Quality 89, Blended $6.00 → Value = 89² ÷ 6 = 1,320

Higher = more quality per dollar.

Context Window

Maximum tokens (input + output) the model can handle in one request.

128K ≈ a 300-page book
1M ≈ a full codebase or ~10 books
10M ≈ massive document corpora

Bigger isn't always better — quality often degrades with very long inputs.

Input Modalities

text — accepts text prompts
img — accepts images (vision)
audio — accepts audio input

Most models are text-only. Multimodal models can analyze images, screenshots, documents.