LLM Observability Pricing: Examples & Companies

What is it

LLM Observability Pricing is pricing for platforms purpose-built to observe, debug, and optimize LLM application behavior — logging prompts, responses, latency, and cost.

The category exists because standard logging and APM tools don’t understand the semantic structure of an LLM interaction. A raw HTTP log of a GPT-4 call shows bytes in and bytes out; an LLM observability platform surfaces the prompt text, the completion, the token counts, the model version, the cost estimate, and any custom metadata (user ID, session, feature flag) the developer tagged the call with. That semantic layer is what enables prompt-engineering workflows, cost attribution by feature, and quality evaluation on production traffic.

Helicone and Portkey are the canonical in-corpus examples of the tight proxy-and-gateway niche — both bill on log volume. It is distinct from the broader observability platform pricing cohort (Langfuse, Arize, Braintrust) that meters traces, spans, and evaluation runs rather than proxied LLM calls. For the underlying question of which unit to bill on, see choosing the right usage metric and understanding usage-based pricing models.

How it works

LLM observability platforms intercept LLM API traffic at the network layer. The customer replaces the provider’s base URL with the observability platform’s proxy URL — a one-line configuration change. Every request then flows through the proxy, which captures the full request/response payload, logs it with metadata, and forwards it to the provider. The billable unit is the log (the recorded call), and pricing follows the pipe: a free log allowance, a flat plan floor, then usage-based overage above the included quota.

The unit math for log-volume pricing:

Monthly cost = plan_fee + max(0, logs − included_allowance) × overage_rate

Worked example on Portkey Production: a 1M-request/month app pays the $49 base plus overage on the 900K requests above the 100K included allowance — nine increments of 100K at $9 each = $81 — for roughly $130/month. At 3M requests the self-serve ceiling caps the bill near $310/month, above which pricing routes to a custom Enterprise contract.

Helicone structures overage differently: Pro ($79) and Team ($799) each include the same 10K requests + 1 GB as the free Hobby tier, then meter overage on logs and storage. Helicone doesn’t publish per-unit overage rates — only an on-page calculator, which estimates roughly $0.97/month for 10K requests with light storage. Both vendors keep the core gateway free and open-source (Helicone Apache 2.0, Portkey MIT), so the paid tiers sell hosted convenience, longer retention, and — at Portkey’s Enterprise tier — active governance like budget caps and rate limits.

Companies using this

Two companies in the corpus are purpose-built LLM observability proxies: Helicone, with a clean gateway model focused on prompt analytics and cost attribution, and Portkey, which pairs observability with active model routing, guardrails, and spend governance. Both price primarily on log volume with generous free tiers, and both were acquired in 2026 (Helicone by Mintlify, Portkey by Palo Alto Networks) — a signal that standalone LLM observability is consolidating into larger platforms.

Explore this theme in the knowledge graph →