What does a finished task cost?
Benchmarks measure models. You ship tasks. So this is a number that matters too: the end-to-end cost to take a real coding job from prompt to a passing result with Sarvam Code.
| Task | Cost | Time | Result |
|---|---|---|---|
| Add a REST API endpoint | ~$0.03 | ~2m | Tests pass |
| Fix a failing test | ~$0.02 | ~1m | Green |
| Explain an unfamiliar repo | ~$0.04 | ~2m | Summary + map |
| Write unit tests for a module | ~$0.05 | ~3m | Tests pass |
| Refactor a module | ~$0.06 | ~4m | Tests pass |
| Add a CLI command | ~$0.04 | ~3m | Builds + runs |
| Generate docs from code | ~$0.03 | ~2m | Docs written |
| Convert JS to TypeScript | ~$0.05 | ~3m | Typechecks |
| Debug a build error | ~$0.04 | ~3m | Builds clean |
| Small full-stack feature | ~$0.12 | ~8m | Feature shipped |
IllustrativePending published runs. Each task will be measured end to end, from prompt to passing result, and the numbers swapped in.
Cost is the model spend for the run. A day of this kind of work stays well inside a single $5 month. See the per-token economics for why.
Keep exploring
Begin here
A gateway between people and intelligence. Join the early-access list and be first to ship with Sarvam Code.
$curl -fsSL https://dl.sarvamcode.com/install.sh | sh
Currently available for macOS only