Best of / AI models / Cheapest LLM API with prompt caching
Best of · AI API prices · checked Oct 23, 2026
Cheapest LLM API with prompt caching
Short answerClaude Haiku 5.5 (Anthropic) is the cheapest: $2.30 per 1,000 cached agent steps. Next: GPT-6 Luna at $2.30 and GPT-5.6 Luna at $5.00. 26 current models qualify (prices checked Oct 23, 2026). Price only, not answer quality.
Every model that qualifies, cheapest first
| # | Model | Provider | Per 1,000 cached agent steps | Context | Input / output per 1M |
|---|---|---|---|---|---|
| 1 | Claude Haiku 5.5 | Anthropic | $2.30 | 1M | $0.1 / $0.5 |
| 2 | GPT-6 Luna | OpenAI | $2.30 | 1.05M | $0.1 / $0.5 |
| 3 | GPT-5.6 Luna | OpenAI | $5.00 | 1.05M | $0.2 / $1.2 |
| 4 | DeepSeek Flash | DeepSeek | $5.58 | 1M | $0.3 / $1.2 |
| 5 | Gemini 3.1 Flash-Lite | $6.25 | 1.05M | $0.25 / $1.5 | |
| 6 | Gemini 3.5 Flash-Lite | $8.90 | 1.05M | $0.3 / $2.5 | |
| 7 | Gemini 3 Flash Preview | $12.50 | 1.05M | $0.5 / $3 | |
| 8 | Gemini 3.8 Flash | $17.25 | 1.05M | $0.75 / $3.75 | |
| 9 | Gemini 3.6 Flash | $17.25 | 1.05M | $0.75 / $3.75 | |
| 10 | GPT-5.4 mini | OpenAI | $18.75 | 400K | $0.75 / $4.5 |
| 11 | DeepSeek V4 Pro | DeepSeek | $22.44 | 1M | $1.32 / $3.96 |
| 12 | Claude Sonnet 5.5 | Anthropic | $43.00 | 1M | $2 / $10 |
| 13 | GPT-6.1 Sol | OpenAI | $43.00 | 1.05M | $2 / $10 |
| 14 | Gemini 4 Argon | $43.00 | 1.05M | $2 / $10 | |
| 15 | GPT-6 Sol | OpenAI | $46.00 | 1.05M | $2 / $10 |
| 16 | GPT-5.6 Terra | OpenAI | $50.00 | 1.05M | $2 / $12 |
| 17 | Gemini 3.1 Pro Preview | $50.00 | 1.05M | $2 / $12 | |
| 18 | GPT-5.2 | OpenAI | $50.75 | 400K | $1.75 / $14 |
| 19 | GPT-5.4 | OpenAI | $62.50 | 1.05M | $2.5 / $15 |
| 20 | Claude Opus 5.5 | Anthropic | $86.00 | 1M | $4 / $20 |
Rules
Current models with a listed cache-read price, ranked by the cost of one coding-agent step (40,000 input tokens, 30,000 of them cached, 2,000 out); cache writes not counted. Current = not deprecated, legacy or shut down on the provider's model page. Prices from each provider's pricing page; full method on LLM prices.
Want a pick for your exact job and volume? Pick for me, or see the cheapest model for each job.
Up next
Another best-of list: Same prices, a different requirement.Cheapest LLM API with a batch discountRead it next →