Skip to main content

Cost analytics

Splyntra turns the token counts on your model spans into money, then rolls that spend up so you can answer the questions that actually matter: what did this run cost, which model dominates our bill, and which project is over budget?

From spans to cost​

Every model span carries the model name and its input/output token counts. Splyntra applies a pricing table for each model to compute a per-span cost, then aggregates:

  • Per run — the total cost of a single agent execution, so you can spot expensive outliers.
  • Per model — spend grouped by model (e.g. gpt-4o vs claude-sonnet vs a local model), useful for right-sizing.
  • Per project — the rollup that maps to a team, product, or environment.
Project: support-agent           (last 7 days)
├─ gpt-4o            18,204 runs   $241.18
├─ text-embedding-3     —          $12.04
└─ gpt-4o-mini       6,910 runs     $8.77
                                  ────────
                          total    $261.99

Because cost lives on the same spans as traces and risk, you can sort runs by cost and open the trace to see exactly which model call or retry drove the number up.

Events and spans​

Cost is derived from the span stream you already send — there is no separate metering integration. Token usage reported via the OpenTelemetry GenAI attributes (gen_ai.usage.input_tokens, gen_ai.usage.output_tokens) is used directly. For providers that don't report usage, Splyntra estimates tokens from the prompt/completion so runs are never uncosted.

Budgets​

You can set a budget per project (and, on Team/Enterprise plans, per environment):

  • Soft budget — an alert fires when spend crosses a threshold (e.g. 80% of the monthly cap).
  • Hard budget — combined with governance spend controls, runs can be blocked or routed for approval when a cap is reached.
tip

Pair budgets with per-model breakdowns to find cheap wins — swapping a summarization step from a frontier model to a smaller one is often invisible in quality but large in cost.

Next steps​

  • Governance — spend controls and approvals.
  • Cloud — plan limits and usage-based billing.