LLM pricing directory
Model prices, per million tokens.
The same price table ReasonRank uses to compute verified savings. As of 2026-07-02.
bedrock
anthropic.claude-3-5-haiku$0.8 / $4anthropic.claude-3-5-sonnet$3 / $15anthropic.claude-3-haiku$0.25 / $1.25anthropic.claude-3-opus$15 / $75anthropic.claude-haiku-4-5$1 / $5anthropic.claude-opus-4-8$5 / $25anthropic.claude-sonnet-4-6$3 / $15anthropic.claude-sonnet-5$3 / $15meta.llama3-1-70b$0.72 / $0.72meta.llama3-1-8b$0.22 / $0.22meta.llama3-3-70b$0.72 / $0.72
anthropic
claude-3-5-haiku$0.8 / $4claude-3-5-sonnet$3 / $15claude-3-7-sonnet$3 / $15claude-3-haiku$0.25 / $1.25claude-fable-5$10 / $50claude-haiku-4-5$1 / $5claude-opus-4$15 / $75claude-opus-4-5$5 / $25claude-opus-4-6$5 / $25claude-opus-4-7$5 / $25claude-opus-4-8$5 / $25claude-sonnet-4$3 / $15claude-sonnet-4-6$3 / $15claude-sonnet-5$3 / $15
gemini-1.5-flash$0.075 / $0.3gemini-1.5-pro$1.25 / $5gemini-2.0-flash$0.1 / $0.4gemini-2.0-flash-lite$0.075 / $0.3gemini-2.5-flash$0.3 / $2.5gemini-2.5-flash-lite$0.1 / $0.4gemini-2.5-pro$1.25 / $10gemini-3-flash$0.5 / $3gemini-3-pro$2 / $12gemini-3.1-flash-lite$0.25 / $1.5gemini-3.1-pro$2 / $12gemini-3.5-flash$1.5 / $9
openai
gpt-4.1$2 / $8gpt-4.1-mini$0.4 / $1.6gpt-4.1-nano$0.1 / $0.4gpt-4o$2.5 / $10gpt-4o-mini$0.15 / $0.6gpt-5$1.25 / $10gpt-5-mini$0.25 / $2gpt-5-nano$0.05 / $0.4gpt-5.1$1.25 / $10gpt-5.2$1.75 / $14gpt-5.3-codex$1.75 / $14gpt-5.4$2.5 / $15gpt-5.4-mini$0.75 / $4.5gpt-5.4-nano$0.2 / $1.25gpt-5.4-pro$30 / $180gpt-5.5$5 / $30gpt-5.5-pro$30 / $180o3$2 / $8o3-mini$1.1 / $4.4o4-mini$1.1 / $4.4
Popular comparisons
claude-haiku-4-5 vs gpt-5.4-minigemini-3-flash vs gpt-5.4-minigpt-5.4-mini vs gpt-5.4-nanogemini-3.1-flash-lite vs gpt-5.4-miniclaude-haiku-4-5 vs gemini-3-flashclaude-haiku-4-5 vs gpt-5.4-nanoclaude-haiku-4-5 vs gemini-3.1-flash-litegemini-3-flash vs gpt-5.4-nanogemini-3-flash vs gemini-3.1-flash-litegemini-3.1-flash-lite vs gpt-5.4-nanoclaude-sonnet-5 vs gpt-5.4gemini-3.5-flash vs gpt-5.4
List prices tell you what a call costs — not whether the cheaper model holds quality on your workload. Run the comparison on your own test cases and get a verified answer.