What does Both Hemispheres sell?
The current public offer focuses on two things: a scalable SAT solver workflow for DIMACS CNF inputs and an OpenAI-compatible LLM API on a shared prepaid balance.
FAQ
Questions buyers usually ask before trying the service
This page is meant to reduce evaluation friction. It covers the public pricing model, current operational limits, and the fastest way to get help from a human.
Answers
Public-service FAQ
What SAT input format is accepted?
The public SAT workflow currently accepts `.cnf` files in DIMACS CNF format up to `100 MB`.
Do you charge for UNSAT or UNKNOWN results?
No. `UNSATISFIABLE` and `UNKNOWN` SAT outcomes are visible for free. The public SAT charge applies when a satisfying assignment exists and you choose to unlock it.
How is SAT pricing calculated?
SAT unlock pricing is based on the number of distinct absolute variable IDs actually used in clauses, not just the DIMACS `p cnf` header: `$10` per `10,000,000` distinct variables. The charge only applies when a satisfying assignment exists and you choose to unlock it.
What is the LLM API shape?
The LLM API is OpenAI-compatible and exposes routes such as `/v1/models` and `/v1/chat/completions`. It also exposes async batch routes for queued chat and embeddings work.
How is theorem pricing calculated?
The public theorem harness currently includes `20` solver-backed runs per customer per month, then charges `$0.05` per additional solver-backed run. The theorem lane still uses the hosted LLM for translation and explanation, so those token charges continue to be billed on the normal LLM lane.
How do trial credits work?
Create an account on the main page, receive the current trial balance, and use it to test the public SAT or LLM paths before topping up.
Do you offer reserved capacity or dedicated deployments?
Yes. The account area now includes a request workflow for `reserved_capacity` and `dedicated_deployment`. Those are reviewed manually and quoted as monthly offers, and accepted deals can now be provisioned as private model aliases on the existing API with enforced included monthly token budgets plus either metered overage or hard blocking. In the first iteration, those private aliases can also be routed to a dedicated upstream endpoint when the deal needs stronger traffic isolation.
Can I request a custom benchmark?
Yes. Email support or submit a capacity request with the workload description and the goal of the benchmark, and a custom evaluation path can be discussed.
Next Step
Read the FAQ, then run one real test
If the answers above fit your use case, use the trial-credit path and benchmark the service on your own prompt flow or CNF instance.