Datadog LLM Observability Pricing
Datadog LLM Observability pricing covers every span your agent logs. It never covers what your customer got for the money.
Datadog markets this product as Agent Observability today, though its own docs and most of the existing writeups about it still call it LLM Observability, which is why search results for its pricing look a little scattered. Strip that naming confusion away and the real numbers are simple: a Free plan at $0 a month covering 40,000 LLM spans with 15 day retention, and a Pro plan at $160 a month covering 100,000 LLM spans at the same retention, both verified against Datadog's own current product page rather than a guess. Only one span type is ever billed, a direct call to an LLM provider; every tool, workflow, agent, embedding, and retrieval span an agent generates along the way is free. This page covers those numbers in full, what happens once an agent runs past them, and the one question Datadog's own dashboards, like every other agent monitoring tool, never answer: what does any of it mean to the customer paying for the agent.
Verified against Datadog's own product page
What Datadog LLM Observability actually costs
Datadog LLM Observability, the product Datadog itself now brands Agent Observability, traces the model calls a live agent makes, alongside the tool calls, reasoning steps, and retrieval steps that surround each one. Billing tracks only one of those event types: an LLM span, defined on Datadog's own pricing page as a single call to an LLM provider. Every tool span, workflow span, agent span, embedding span, and retrieval span an agent produces around that one model call is free, no matter how many of them a single customer message triggers. The Free plan costs $0 a month for up to 40,000 LLM spans a month with 15 day retention and full feature access, evaluations included. The Pro plan costs $160 a month for up to 100,000 LLM spans at the same 15 day retention, with no separate charge for evaluations there either. Both figures come straight from Datadog's current product page, not a list price someone forgot to update.
Past the Pro plan's 100,000 span allowance, Datadog keeps tracing rather than cutting an agent off: additional usage bills per 10,000 LLM spans, though Datadog doesn't publish the exact per unit rate on the page itself, only that month to month and annual commitments get a discount against it. Retention follows the same per-span logic. The default 15 days can extend to 30, 60, or 90 days for traces and 6, 9, or 12 months for experiments, each also billed per 10,000 spans rather than folded into the base plan price. Practically, that makes Datadog LLM Observability the strongest fit for a team that already pays for Datadog's infrastructure and APM products and wants agent tracing sitting inside the same dashboards a platform team already watches, rather than a first tool for a team with no existing Datadog footprint at all.
Four numbers that matter
The four numbers that decide the real bill
Straight from Datadog's own product page, not a third party estimate.
Free plan: $0 a month
Covers up to 40,000 LLM spans a month with 15 day retention and full feature access, evaluations included. No credit card charge until an agent outgrows it.
Pro plan: $160 a month
Covers up to 100,000 LLM spans a month at the same 15 day retention. The step almost every team past the testing stage ends up taking.
What actually gets billed
Only a direct call to an LLM provider counts as a billable span. Tool, workflow, agent, embedding, and retrieval spans around it are free, regardless of volume.
Retention add-ons
Trace retention extends to 30, 60, or 90 days; experiment retention to 6, 9, or 12 months. Both billed per 10,000 spans, discounted on annual commitments.
What the span counter doesn't show
Datadog tells your engineers what a span cost. It tells your customer nothing
Every dashboard Datadog LLM Observability produces, the span counts, the retention settings, the climbing bill past 100,000 spans a month, renders inside a workspace scoped to your own Datadog account, read by your own engineers. That is true of the whole agent monitoring tool category, not just Datadog: Langfuse, Galileo, Fiddler, and Helicone all share the same design, a single account with no notion of a paying customer as a separate audience who might need their own scoped view of anything. A span counter answers whether an agent is staying inside its quota. It says nothing about deflection rate, the share of conversations an agent resolved without a human, or about cost per resolution scoped to one specific customer's own traffic.
If you sell an AI agent inside a product other companies pay for, that second question is the one your own customer actually asks once the agent is live and the renewal conversation starts. AiAgRe's Node SDK reads the same kind of agent events a tracer like Datadog already produces, tags each one with an organization identity and a customer identity at the point of ingestion, and turns that into white-labeled dashboard components a customer sees inside your own product, scoped so tightly that one customer's numbers never reach another customer's view. It doesn't replace Datadog's span level tracing or its infrastructure monitoring, and most teams run both: Datadog for what the agent did, AiAgRe for what that activity was worth to the person paying for it.
FAQs
Datadog LLM Observability pricing: frequently asked questions
Common questions from teams pricing Datadog's agent tracing against the rest of the market before they commit to it.
How much does Datadog LLM Observability actually cost?
The Free plan costs $0 a month and covers up to 40,000 LLM spans with 15 day retention and full feature access, including evaluations. The Pro plan costs $160 a month for up to 100,000 LLM spans, also with 15 day retention. Past that 100,000 span allowance, Datadog bills additional on demand usage per 10,000 LLM spans, and it doesn't publish the exact per unit rate on its own pricing page, so a team past the Pro plan's included quota needs to talk to Datadog sales to see the real number. These figures come from Datadog's own product page as of this page's publish date and are worth reconfirming there directly before budgeting against them, since usage based pricing changes.
What counts as a billable LLM span in Datadog?
Only one span type gets billed: a single call to an LLM provider, the actual request that goes out to a model like GPT or Claude and the response that comes back. Tool spans, workflow spans, agent spans, embedding spans, and retrieval spans are all free, no matter how many of them a single agent turn generates. That matters more than it sounds like it should, because a single customer message to an agent can easily produce a dozen tool calls and reasoning steps for every one actual model call, so the real bill tracks the number of model calls an agent makes, not the number of steps it takes to get there.
Is Datadog LLM Observability worth it if I don't already use Datadog?
Usually not as a first choice. Datadog LLM Observability is a line item inside a much larger observability platform built for infrastructure and APM, and it earns its keep fastest for a team that already pays for Datadog elsewhere and wants agent tracing inside the same dashboards its platform team already watches, rather than a separate login and a separate bill. A team with no existing Datadog footprint usually finds a dedicated tracer like Langfuse, or an evaluation first tool like Galileo or Fiddler, faster to adopt and cheaper to start with. The fuller breakdown of those alternatives, what each one costs and what each one is actually built for, is on the AI agent monitoring tools comparison.
Does Datadog LLM Observability show anything to my own customers?
No, and that gap is consistent across the whole category, not unique to Datadog. Every dashboard, every span, every cost figure Datadog LLM Observability produces renders inside a workspace scoped to your own Datadog account, built for your own engineers to read. There is no concept of a paying customer as a separate audience with their own scoped view, the same gap every AI agent monitoring tool on the market shares today. If you sell an AI agent inside a product other companies pay for, your own customer wants their own deflection rate and cost per resolution, isolated to their own traffic only, which is a different job than watching a span counter climb toward a quota.
What happens after an agent goes over the Pro plan's span limit?
Datadog keeps tracing every LLM span past the 100,000 included in the Pro plan; it just starts billing the overage on a per 10,000 span basis rather than cutting the agent off. Retention can also be extended past the default 15 days, out to 30, 60, or 90 days for traces and 6, 9, or 12 months for experiments, each also billed per 10,000 spans, with Datadog offering a discount on month to month and annual commitments. None of those add on rates are published as a flat number either, so a team running real production volume should treat the $0 and $160 tiers as a starting point, not the ceiling on the real bill.
Can I run AiAgRe alongside Datadog LLM Observability?
Yes, and that is the normal setup rather than a workaround. AiAgRe's Node SDK reads the same kind of underlying agent events a tracer like Datadog LLM Observability already produces, tags each one with an organization identity and a customer identity at the point of ingestion, and turns that into deflection rate, cost per resolution, and resolution rate shown through white labeled dashboard components inside your own product. Nothing about installing it touches how Datadog traces spans or bills for them; the two run on separate SDKs reading the same agent, one scoped to your engineers and one scoped to your paying customers.
How does Datadog LLM Observability compare to Langfuse or LangSmith?
All three trace the same underlying agent activity, but they start from different places. Langfuse is open source and SDK first, built specifically for LLM and agent tracing from the ground up. LangSmith is LangChain’s own tool, tightest for a team already building on LangChain. Datadog LLM Observability is a module inside a much larger infrastructure and APM platform, the strongest fit once agent tracing needs to sit next to metrics a platform team already watches for the rest of the stack. None of the three differ on the bigger gap this page covers: none of them scope any of it to a paying customer outside your own company.
Already tracing spans in Datadog? Now show your customer what they got.
Request access and we'll walk through how AiAgRe's embed tokens map onto the trace data your tracer already produces.