What is it
Per-trace pricing is a billing unit where each distributed trace — a complete record of an LLM request chain — is metered, common in AI observability platforms serving agentic and multi-step LLM applications.
A trace is a higher-order unit than a log event. Where a log captures a single LLM call, a trace captures an entire agent run: the root prompt, all intermediate reasoning steps, tool calls, retrieval operations, sub-agent delegations, and the final response. LangSmith, its parent company LangChain, and Weights & Biases all meter this trace-shaped data because they serve teams where the meaningful unit of work is the user request, not the individual model call.
The trace unit matters most as AI applications move from single-turn calls to multi-step agent workflows. A customer support agent might make 8–15 LLM calls to resolve one ticket. Per-log pricing charges for each internal call; per-trace pricing charges once per agent run, matching the customer’s intuition of “I processed one request.” That rewards complex chains — a 20-step agent pays the same per-trace rate as a single-call app — so LangSmith and Weights & Biases both ship free tiers sized to developer and small-team volume. For how vendors pick meters like this, see choosing the right usage metric.
How it works
Trace billing opens a trace record when the root span starts and closes it when the final span completes. All child spans (sub-calls, tool invocations, retrievals) nest inside the root trace. The platform stores the full tree and bills the trace regardless of how many spans it contains — though vendors differ on whether they count traces or the bytes those traces ingest.
| Company | Trace unit | Free tier allotment | Overage rate |
|---|---|---|---|
| LangSmith | One run tree per request; two retention classes | 5k base traces/mo (Developer, $0); 10k (Plus, $39/seat/mo) | $2.50 / 1k base (14-day); $5.00 / 1k extended (400-day) |
| LangChain | Same trace meter on the LangSmith platform | 5k (Developer) / 10k (Plus) base traces/mo | $2.50–$5.00 / 1k traces (one of seven meters) |
| Weights & Biases | W&B Weave data ingestion (GB of logged traces) | 1 GB/mo (Free); 1.5 GB/mo (Pro, $60/mo) | $0.10 / MB above the included allotment |
The unit math for LangSmith and LangChain’s countable per-trace model:
Monthly trace cost = max(0, traces − included_allowance) ÷ 1,000 × rate_per_1k
A five-seat team on LangSmith Plus running ~200k traces/month on extended retention lands around $1,145/month — $195 in seats plus roughly $950 in trace overage. Because every LangChain Fleet run ($0.05/run) is also traced by default, trace volume compounds with the other meters on that platform.
Weights & Biases prices the same trace data differently: its W&B Weave observability product meters logged trace ingestion by volume at $0.10/MB (roughly $100/GB) rather than counting individual traces. A GenAI app that logs verbose traces and large prompts can produce several GB a month, so a team that budgeted “$60 for W&B Pro” can see a four-figure ingestion line — trace pricing expressed as data throughput rather than a per-run count.
Companies using this
Three companies in the corpus meter LLM traces, all serving teams where the agent run — not the individual LLM call — is the natural unit of application behavior to track and debug. For how these platforms capture the underlying data, see tracking and metering usage events.
| Company | Product | Pricing model | Billing units | Free tier | Verified |
|---|---|---|---|---|---|
| LangChain | Agent orchestration frameworks + LangSmith platform | Yes | 2026-07-29 | ||
| LangSmith | LLM tracing and evaluation | Yes | 2026-07-29 | ||
| Weights & Biases | MLOps experiment tracking, W&B Weave LLM observability/evals, Models registry, and Serverless Inference | Yes | 2026-07-29 |
Explore this theme in the knowledge graph
FAQ
What is a trace in LLM observability pricing?
A trace is a complete record of one agent or chain run — capturing every step, sub-call, tool invocation, and token count from start to finish. LangSmith bills traces directly (a run tree for one request), while Weights & Biases meters the same data as Weave ingestion volume. One user query to an agent might produce dozens of LLM calls but is a single trace, making it a more meaningful billing unit than individual log events for multi-step workflows.
How much does per-trace pricing cost?
LangSmith charges $2.50 per 1,000 base traces (14-day retention) and $5.00 per 1,000 extended traces (400-day retention), after 5,000 free traces on the Developer plan or 10,000 on the $39/seat Plus plan. LangChain uses the same trace rates on the LangSmith platform. Weights & Biases meters traces as W&B Weave data ingestion instead, charging $0.10/MB of logged trace data above the included 1–1.5 GB per month.
Does per-trace pricing include free traces?
Yes. All three vendors include a free trace allotment. LangSmith and LangChain include 5,000 base traces per month on the free Developer plan and 10,000 on Plus. Weights & Biases includes 1 GB of Weave trace ingestion per month on the free tier and 1.5 GB on Pro before per-MB overage applies.
How does per-trace pricing scale for agentic applications?
Per-trace pricing scales with the number of agent runs, not the internal complexity of each run. A user who sends 1,000 queries generates 1,000 traces whether each run makes 5 LLM calls or 50, so teams running long agent chains get a favorable effective cost per LLM call. The trade-off is that on LangSmith a trace can auto-upgrade from the $2.50/1k base rate to the $5.00/1k extended rate the moment you add feedback, queue it for annotation, or an automation rule matches it.
Related billing units
- Credit-Based BillingA billing unit where customers pre-purchase or are allocated a pool of credits that deplete as they use the product, often at variable rates per feature.
- Token-Based PricingA billing unit common in LLM and AI products, where customers are charged per input and output token processed.
- Per-Seat PricingA billing unit where the vendor charges a fixed fee per named user, regardless of how much each user consumes.
- Per-Resolution PricingA billing unit unique to AI customer-support products, where the vendor charges only when an AI agent resolves a customer issue without escalation.
- Bandwidth-Based PricingA billing unit where customers are charged per gigabyte of data transferred out of the platform.
- Per-Function-Invocation PricingA billing unit where customers are charged per serverless function invocation, often combined with a separate compute-time charge.
- CPU-Hour PricingA billing unit where customers are charged for the CPU time their workloads consume, typically measured in vCPU-seconds or vCPU-hours.
- GB-Hour PricingA billing unit where customers are charged for the memory their workloads consume over time, measured in gigabyte-hours.
- GPU-Hour PricingA billing unit where customers are charged for GPU time consumed, typically measured per-second or per-hour by GPU type.
- Per-API-Call PricingA billing unit where customers are charged per API request, regardless of payload size or processing time.
- Per-GB Storage PricingA billing unit where customers are charged per gigabyte of data stored on the platform per month.
- Media-Minute PricingA billing unit where customers are charged per minute of audio or video processed — used by speech, voice, and video AI vendors.
- Per-Request PricingA billing unit where customers are charged per request served — the generic meter for inference endpoints, search, scraping, and browser infrastructure.
- Per-Event PricingA billing unit where customers are charged per event ingested — the native meter of observability and billing-infrastructure platforms.
- Vector Storage PricingA billing unit where customers are charged for vectors stored or indexed — the storage dimension of vector database pricing.
- Per-Character PricingA billing unit where customers are charged per character of text processed — the standard meter for text-to-speech and translation.
- Per-Document PricingA billing unit where customers are charged per document processed or generated — common in AI writing, SEO, and document-intelligence tools.
- Per-Page PricingA billing unit where customers are charged per page crawled, parsed, or rendered — the meter for web scraping and document parsing.
- Per-Transaction PricingA billing unit where customers are charged per financial or billing transaction processed — the meter of billing and accounting platforms.
- Active-User PricingA billing unit where customers are charged per monthly or daily active user rather than per provisioned seat.
- Per-Task PricingA billing unit where customers are charged per task an automation or agent executes — Zapier's historical unit, now spreading to AI agents.
- Per-Unit PricingA billing unit used by robotics, hardware AI, and some SaaS companies where the metered object is a physical or abstract 'unit' — a robot deployed, a device sold, or a defined deliverable.
- Workflow Execution PricingA billing unit where each end-to-end workflow or automation run is metered and billed, regardless of the compute steps it contains.
- Per-Message PricingA billing unit where each individual message or reply in a conversation is metered, common in AI chat and voice platforms.
- Per-Invoice PricingA billing unit used by billing infrastructure platforms where each invoice generated or processed is metered as the primary cost driver.
- Per-Action PricingA billing unit where each discrete action taken by an AI agent or automation is metered — common in browser automation and agentic workflow tools.
- Per-Image PricingA billing unit where each AI-generated image is metered, common in image generation APIs and multimodal AI platforms.
- Per-Conversation PricingA billing unit where each complete customer conversation — from first message to resolution — is metered as a single chargeable event.
- Per-Record PricingA billing unit where each data record processed, labeled, or extracted is metered — common in data platforms and web scraping services.
- Per-Word PricingA billing unit common in translation and localization platforms where the metered object is the word count of content processed.
- Per-Video PricingA billing unit where each AI-generated video is metered, common in video generation and synthetic media platforms.
- Milestone-Based PricingA billing unit used in drug discovery and biotech AI where payment is tied to achieving defined research milestones rather than time or compute consumed.
- Per-Outcome PricingA billing unit where payment is triggered by verified outcomes delivered — distinct from outcome-based pricing models, this refers specifically to 'outcomes' as a countable billing unit.
- Per-Datapoint PricingA billing unit where each individual data measurement or signal ingested is metered — common in cloud cost intelligence and ML evaluation platforms.
- Per-Interaction PricingA billing unit where each patient-agent or user-agent interaction is metered, common in healthcare AI and customer engagement platforms.
- Data Licensing PricingA pricing structure where access to proprietary datasets or data assets is licensed separately from the software or services, common in AI training data and clinical data platforms.
- Robot-Hour PricingA billing unit where each hour a robot or autonomous system operates is metered — the robotics equivalent of a GPU-hour.
- Per-Contact PricingA billing unit where each contact or lead in the database is metered, common in AI sales development and outbound automation platforms.
- Per-Mailbox PricingA billing unit where each connected email mailbox or sending account is metered, common in AI outbound sales and email automation platforms.
- Browser-Hour PricingA billing unit where each hour of headless browser compute time is metered, common in web scraping and browser automation platforms.
- Per-Generation PricingA billing unit where each AI-generated creative asset — image, video, or design — is counted as a 'generation' and metered accordingly.
- Per-Ticket PricingA billing unit where each customer support ticket handled by an AI agent is metered — common in AI customer service platforms.
- Per-Log PricingA billing unit where each LLM request log ingested or stored is metered — common in AI observability and evaluation platforms.
- Per-IP PricingA billing unit where each IP address or proxy endpoint allocated is metered — used by web scraping proxy providers.
- Per-Device PricingA billing unit where each hardware device or endpoint connected to the AI platform is metered.
- Per-Case PricingA billing unit used in legal AI platforms where each case or matter processed by the AI is metered.
- Per-Report PricingA billing unit where each AI-generated report or analysis document is metered as a discrete output.