workload
Answer questions over your docs
Retrieved chunks plus a question, short answer.
Retrieval pads the prompt, so caching the corpus changes the bill.
- Input per request
- 6k
- Output per request
- 500
- Requests per month
- 20,000
- Tokens per month
- 130M
- Cheapest today
- $2.58
- Most expensive
- $1,700
Most of the spread comes from output tokens on this workload. Applying a 60% prompt cache to the cheapest row moves it by $0 a month.
| Model | In /1M | Out /1M | Per request | Per month | Share | Buy |
|---|---|---|---|---|---|---|
| Mistral NemoMistralcheapestopen weights | $0.019 | $0.030 | $0.000129 | $2.58 | Open | |
| GPT-OSS 20BOpenAIopen weights | $0.018 | $0.090 | $0.000153 | $3.06 | Openaff | |
| Qwen3.7 FlashAlibaba | $0.030 | $0.130 | $0.000245 | $4.90 | Open | |
| Llama 3.1 8BMetaopen weights | $0.050 | $0.080 | $0.00034 | $6.80 | Openaff | |
| DeepSeek V4 FlashDeepSeek | $0.086 | $0.171 | $0.000601 | $12.03 | Open | |
| Gemini 2.5 Flash-LiteGoogle | $0.100 | $0.400 | $0.0008 | $16.00 | Open | |
| GLM 5.3 FlashZ.ai | $0.150 | $0.500 | $0.0012 | $23.00 | Open | |
| Gemini 3.8 FlashGoogle | $0.750 | $3.75 | $0.0064 | $128 | Open | |
| Claude Haiku 4.5Anthropic | $1.00 | $5.00 | $0.0085 | $170 | Openaff | |
| Grok 4.3xAI | $1.25 | $2.50 | $0.0088 | $175 | Open | |
| GPT-5.1OpenAI | $1.25 | $10.00 | $0.0125 | $250 | Open | |
| Grok 4.6xAI | $2.00 | $6.00 | $0.015 | $300 | Open | |
| Claude Sonnet 5Anthropic | $2.00 | $10.00 | $0.017 | $340 | Openaff | |
| GPT-5.2 CodexOpenAI | $1.75 | $14.00 | $0.0175 | $350 | Open | |
| Gemini 3.1 ProGoogle | $2.00 | $12.00 | $0.018 | $360 | Open | |
| GPT-5.4OpenAI | $2.50 | $15.00 | $0.0225 | $450 | Open | |
| Kimi K3Moonshot | $3.00 | $15.00 | $0.0255 | $510 | Open | |
| Claude Opus 5.5Anthropic | $4.00 | $20.00 | $0.034 | $680 | Openaff | |
| GPT-5.5OpenAI | $5.00 | $30.00 | $0.045 | $900 | Open | |
| Claude Fable 5.1Anthropic | $10.00 | $50.00 | $0.085 | $1,700 | Openaff | |
| GPT-6 AstraOpenAI | $10.00 | $50.00 | $0.085 | $1,700 | Open |
Other workloads
Answer support ticketsSummarize documentsReview pull requestsRun an agent over a repoClassify rows at scaleDraft docs and changelogs
Figures assume no retries, no failed calls, and no tool-use tokens. How the arithmetic works.
Prices read from 6 provider pricing pages on Sep 24, 2026. Some outbound links may be affiliate links. How that works.