As of June 5, 2026, a Comet Opik pricing calculator should start with spans and retention instead of generic LLM request counts. Estimate workflow runs, average spans per run, included span allowance, paid span overage, standard retention, whether the 400-day retention extension is needed, and then add evaluation, annotation, LLM-judge, and self-host rows.
The short answer: Opik cost planning is an observability ledger. Hosted Comet pricing covers plan floors, span allowances, added spans, retention, support, and enterprise deployment paths; Makefun still needs separate rows for evaluator model tokens, human review, storage, self-host operations, and WordPress publishing QA.
Official source snapshot
Publisher refreshed the official Comet pricing page, Opik docs hub, tracing overview, evaluation overview, and self-host overview. Treat those pages as the current source of truth because span allowances, retention terms, support rows, and enterprise options can change.
| Worksheet row | Opik input | Makefun caveat |
|---|---|---|
| Trace and span volume | Workflow runs, spans per run, included spans, overage spans, retry traces, and failed agent branches. | Normalize by span shape, not request count, before comparing observability tools. |
| Retention | Default retention, longer audit window, 400-day extension need, trace payload size, and source-refresh cadence. | Longer retention can be an audit requirement, not only a storage preference. |
| Evaluation | Test-suite runs, online-eval sampling, annotation queues, evaluator prompts, LLM-judge tokens, and regression reruns. | Provider model tokens and human review time sit outside the Opik plan row. |
| Self-host | Kubernetes or local deployment, storage, backups, monitoring, upgrades, access control, and security review. | Self-hosting trades vendor plan spend for operations and governance work. |
Span and retention formula
monthly_spans = workflow_runs * average_spans_per_run + retry_spans + failed_branch_spans + eval_trace_spans. Compare that total with the current included allowance, then add overage rows only after publication-day pricing refresh.
- Count source refresh, support-agent replies, media QA, WordPress publish gates, and coding-agent maintenance as separate workflow lanes.
- Keep trace ingestion, retention extension, online evals, annotation work, and evaluator model spend in different rows.
- Use official pricing for current plan floors and span rows; avoid hard dollar claims when a row is enterprise-only or customized.
- Do not claim Opik is cheaper, better, more secure, or more reliable without same-workflow evidence.
Evaluation and annotation rows
| Scenario | Cost driver | Control |
|---|---|---|
| SEO source refresh | Official-source trace spans, retries, evaluator prompts, and failed-source regression tests. | Sample only high-risk source changes and separate model-token spend from Opik spans. |
| Support-agent QA | Conversation spans, feedback scores, annotation queues, and low-confidence review loops. | Track reviewer time and customer-risk severity outside the hosted plan. |
| Media QA | Image or video metadata traces, generation retries, checklist spans, and editorial acceptance labels. | Do not store sensitive payloads longer than the audit window needs. |
| Coding-agent maintenance | Tool-call traces, dataset rows, regression evals, and release-review annotations. | Use retention extension only where post-release debugging requires it. |
Same-use comparison lanes
Use adjacent Makefun calculators as comparison lanes, not same-intent duplicates: Helicone Gateway Cache Session Cost Calculator for gateway observability, Vellum Workflow Evaluation Release Review Cost Calculator for workflow eval release review, Requesty AI Gateway Cache Routing Cost Calculator for gateway routing and cache costs, Botpress AI Spend Message Event Knowledge Cost Calculator for support-agent event spend, Upstash QStash Message Retry Workflow Cost Calculator for retry workflow gates, Pinecone Assistant Context Token RAG Cost Calculator for RAG context-token planning, and Cohere Rerank Model Vault Private Search Cost Calculator for rerank evaluation scenarios.
Langfuse, Braintrust, Helicone, and Phoenix pricing pages were refreshed as same-use references. Their units differ from Opik spans, so the comparison should normalize the same workflow before discussing relative cost. Braintrust, Langfuse, and LangSmith Makefun candidate pages were not used as body links unless a live 200 page was already verified.
Publisher checklist
- Refresh Comet pricing, Opik tracing, evaluation, and self-host docs before reusing this calculator.
- Verify target URL, WordPress search, sitemap, Yoast fields, blog category id 2, body links, and unique permanent media before a success notification.
- Keep evaluator model tokens, storage, vector database work, infrastructure, and human annotation outside the Opik plan row.
- Do not embed Comet or Opik logos, copied UI, copied pricing tables, third-party SaaS logos, generated people, temporary URLs, or media ID 7037.
FAQ
What should a Comet Opik pricing calculator include? Include workflow runs, average spans per run, included spans, overage spans, retention days, retention extension, test-suite executions, online-eval sampling, annotation items, LLM-judge token spend, hosted plan floor, and self-host infrastructure work.
Is Opik cheaper than Langfuse, Braintrust, or LangSmith? Do not make a blanket cheapest claim. Normalize the same workflow first because Opik spans, Langfuse units, Braintrust processed data and scores, LangSmith traces, Helicone requests, Phoenix spans or GB, and self-host logs use different meters.
When should Makefun consider self-hosted Opik? Consider self-hosting only when private trace payloads, custom retention, deployment control, or infrastructure policy outweigh Kubernetes, storage, backup, monitoring, upgrade, and access-control work.
