API prices per million tokens
| Model | Input | Output |
|---|---|---|
| Claude Opus 5.5 | $4 | $20 |
| GPT-5.6 Terra Pro | $2 | $12 |
| Gemini 3.1 Pro | $2 | $12 |
| Claude Sonnet 5.5 | $2 | $10 |
| Grok 4.7 | $2 | $6 |
| DeepSeek V4 Flash (open, unchecked) | $0.03 | $1.28 |
US dollars per million tokens, list prices on OpenRouter checked 2026-10-07. Thinking tokens are billed as output.
How the numbers are worked out
A token is about three quarters of a word, so a 10-page contract of 5,000 words is about 6,700 input tokens. Output tokens are the answer the model writes, plus its thinking when it thinks; frontier models usually spend more on output than on input for document work, which is why the output price matters.
Yellowjacket's price is 35% of the cheapest top model's estimate for the same job, shown before the job runs, so you keep at least 65% whichever model you would have used. Our own cost is lower still: on 80 real contracts Opus alone cost $4.41 and Yellowjacket's models cost $0.50.
Questions people ask
What is the cheapest LLM API?
Per token, small open models such as DeepSeek V4 Flash cost about $0.03 per million input tokens, over 100 times less than a frontier model. On their own they make more mistakes; Yellowjacket uses them for reading and checks every answer, escalating the misses.
How much does the Claude API cost?
Claude Opus 5.5 is $4 per million input tokens and $20 per million output tokens; Claude Sonnet 5.5 is $2 and $10. A 10-page contract with a one-page answer costs about 3-6 cents on Opus.
How much does the OpenAI API cost?
GPT-5.6 Terra Pro is listed at $2 per million input tokens and $12 per million output tokens (OpenRouter list price, October 2026).
Why is my AI bill higher than the calculator says?
Thinking tokens, retries and agents that re-read the same files. Agents are where Yellowjacket saves most: an invoice exception desk cost $0.79 per 30 invoices as an Opus agent and $0.006 per batch on Yellowjacket once the plan existed.