Loading pricing data…
Compare prices across 1437 models from 21 providers
| Compare | Provider | Model | Input Price | Output Price |
|---|---|---|---|---|
| Anthropic | $10 | $50 | ||
| Anthropic | $0.8 | $4 | ||
| Anthropic | $0.8 | $4 | ||
| Anthropic | $1 | $5 | ||
| Anthropic | $10 | $50 | ||
| Anthropic | $15 | $75 | ||
| Anthropic | $15 | $75 | ||
| Anthropic | $15 | $75 | ||
| Anthropic | $15 | $75 | ||
| Anthropic | $15 | $75 | ||
| Anthropic | $5 | $25 | ||
| Anthropic | $5 | $25 | ||
| Anthropic | $5 | $25 | ||
| Anthropic | $5 | $25 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $3 | $15 | ||
| Anthropic | $2 | $10 | ||
| AWS Bedrock | $0.3 | — | ||
| AWS Bedrock | $0.3 | — | ||
| AWS Bedrock | $1.25 | — | ||
| AWS Bedrock | $3 | — | ||
| AWS Bedrock | $0.3 | — | ||
| AWS Bedrock | $0.04/image | — | ||
| AWS Bedrock | $0.06 | — | ||
| AWS Bedrock | $0.035 | — | ||
| AWS Bedrock | $0.0675 {priceOf!bedrock/bedrock!fWTU_nxHvSmnGvjH1erX6tTZZ— | — | ||
| AWS Bedrock | $0.135 {priceOf!bedrock/bedrock!Zfo8texvBEL9ZYfvvuztxH_ja— | — | ||
| AWS Bedrock | $2.5 | — | ||
| AWS Bedrock | $0.8 | — | ||
| AWS Bedrock | $1 | — | ||
| AWS Bedrock | $0.08/video | — | ||
| AWS Bedrock | $3.4 | — | ||
| AWS Bedrock | $0.005/image | — | ||
| AWS Bedrock | $0.008/image | — | ||
| AWS Bedrock | $0.008/image | — | ||
| AWS Bedrock | $0.0002/image | — | ||
| AWS Bedrock | $0.1 | — | ||
| AWS Bedrock | $0.02 | — | ||
| AWS Bedrock | $0.2 | — | ||
| AWS Bedrock | $0.15 | — | ||
| AWS Bedrock | $0.5 | — | ||
| AWS Bedrock | $8 | $24 | ||
| AWS Bedrock | $8 | $24 | ||
| AWS Bedrock | $0.25 | $1.25 | ||
| AWS Bedrock | $15 | $75 | ||
| AWS Bedrock | $3 | $15 |
per 1M tokens · US Dollar · Source: Official pricing pages
Mistral OCR (25.05) (Google Vertex AI) currently offers the lowest input price at $0.0005 per million tokens for text generation.
Meta: Llama 4 Scout (OpenRouter) supports the largest context window among tracked models at 10M.
Most LLM APIs charge separately for input tokens (your prompt and conversation history) and output tokens (the model's response). Output tokens typically cost 2–5× more than input tokens. Many providers also offer discounted batch pricing, prompt caching for repeated prefixes, and surcharges for long-context requests.
LLMRates.ai tracks models from Anthropic, Google, OpenAI, AWS Bedrock, Azure AI, Cohere, Cursor, DeepSeek, and more — covering both US and Chinese AI providers.
Prices are scraped directly from official provider pricing pages and updated daily. Major price changes are typically reflected within 24 hours.
Prompt caching stores a repeated prompt prefix — such as a system prompt or a large document — so the model doesn't re-process it on every request. Providers like Anthropic and Google offer cached input tokens at a 70–90% discount off the standard input price, making it a key cost-reduction lever for production applications.
gpt-realtime-1.5 (OpenAI) costs $32 per million input tokens and $64 per million output tokens.
Claude Sonnet 5through August 31, 2026 (Anthropic) costs $2 per million input tokens and $10 per million output tokens.
Gemini Omni Flash Preview (Google) costs $1.5 per million input tokens and $9 per million output tokens.
The most affordable popular models by input price (USD, standard tier, per million tokens):
AI agents run in multi-step loops — each iteration re-sends the accumulated conversation history, tool results, and plan, so token costs compound quickly. Prioritize models with low input pricing and strong tool/function-calling support. Prompt caching (Anthropic, Google) can cut costs by 70–90% on repeated context. Affordable models with solid tool support include:
Reasoning models — such as glm-4.5, glm-4.5-air, glm-4.6, glm-4.7 — perform internal chain-of-thought reasoning before responding. They are typically priced 5–20× higher than standard models due to the extra compute for reasoning tokens. Some providers charge a separate rate for thinking tokens generated during the reasoning process.