Resources / Compare / LLM cost tracking tools
COMPARE · COST TRACKING TOOLS
Seven tools that show what your LLM calls cost, grouped by how they work: observability platforms, gateways, and cost-only trackers. Each entry says what it breaks spend down by, what it stores, and what it charges, checked against its own pages. LLMtrack is one of them, and it does AI cost tracking per feature, model and customer.
Quick answer
Pick by the question you need answered. Debugging what a chain did: Langfuse or LangSmith. Routing, caching and spend caps across providers: a gateway such as LiteLLM, Cloudflare AI Gateway or Portkey. Knowing which feature and which customer spends the money, and whether that customer is profitable: LLMtrack, alone or next to either of the others.
OBSERVABILITY AND GATEWAYS
COST-ONLY TRACKERS
Many teams run one of each. The table below shows where each tool sits.
| Tool | Type | Cost broken down by | Free plan | Paid from |
|---|---|---|---|---|
| Langfuse | Observability, open source | Model, tags, users | 50k units/mo, 30 days | Core $29/mo |
| LangSmith | Observability | Not covered here | 5k base traces/mo, 14-day retention | Plus $39/seat/mo |
| Helicone | Gateway, maintenance mode | Not listed on its pricing page | 10k requests, 7-day retention | Pro $79/mo + usage |
| Portkey | Gateway | Request, model, provider, key | 10k logs/mo, 3-day log retention | Production $49/mo |
| LiteLLM | Self-hosted proxy | Key, user, team; tags on enterprise | You run it | Enterprise license for some features |
| Cloudflare AI Gateway | Hosted gateway | Analytics incl. cost | Available on all plans | See Cloudflare's plans |
| LLMtrack | Cost-only tracker | Feature, model, customer, margin | 3,000 requests/mo, 30 days, every feature | Pro $11.99/mo |
From each product's own pricing page or docs, 16 Oct 2026. "Not covered here" means we didn't verify it for this page. Units, traces, logs and requests are each counted differently, so the free tiers aren't directly comparable.
An LLM engineering platform with tracing, evals and prompt management. Its docs say it breaks cost down by model, tags and users, and prices cached, audio and reasoning tokens as separate usage types. Open source and self-hostable; it joined ClickHouse and says its cloud "keeps running as-is".
Best for: teams that debug with traces and want cost in the same place. Compared in detail: Langfuse alternative for AI cost tracking.
LangChain's platform for tracing and evaluating LLM apps. The free Developer plan includes 5k base traces a month with 14-day retention; Plus is $39 per seat a month.
Best for: teams building on LangChain who want tracing and evals from the same vendor.
A gateway you reach by changing your base URL, with request logs, caching, rate limits and fallbacks. Helicone joined Mintlify and has been in maintenance mode since March 2026, with security patches and new-model support still shipping.
Best for: existing users whose caching or fallbacks are doing real work. Leaving? See the Helicone alternative for LLM cost tracking.
A gateway with per-request cost tracking and budget limits. Its docs say budgets apply to models it has pricing for, and that requests to models without pricing show $0.00. Now part of Palo Alto Networks, under the Prisma AIRS AI Gateway name.
Best for: teams that want spend caps enforced in the request path. Small teams weighing a move: Portkey alternatives for small teams.
A proxy you run yourself. Its docs say it tracks spend "for keys, users, and teams across 100+ LLMs"; spend by tag is an enterprise feature.
Best for: teams that want the gateway on their own servers. It connects to LLMtrack with one URL: LiteLLM cost tracking.
Caching, rate limiting, retries, model fallback, logs, and analytics that include cost. Cloudflare lists it as available on all plans.
Best for: apps already on Cloudflare. It connects to LLMtrack with one URL: Cloudflare AI Gateway cost tracking.
One async call after each response, or one URL in your gateway. Each request is priced component by component from the AI API pricing sheet for 422 models, then broken down by feature, model and customer, with margin against what each customer pays, a forecast with the day you pass your budget, and model-switch advice. Unknown models are flagged as unpriced, never shown as $0. Prompts are never sent.
What it doesn't do: tracing, routing or caching. Pair it with one of the tools above if you need those. Install the Node and Python cost tracking SDKs, or estimate a feature first with the AI API cost calculator.
ASK FIRST
THEN CHECK
For the setup side, the guide to tracking LLM costs in production covers naming features and customers so any of these tools gives a useful breakdown.
What is the best LLM cost tracking tool?
It depends on the question. For traces with cost attached, Langfuse or LangSmith. For spend control in the request path, a gateway such as Portkey, LiteLLM or Cloudflare AI Gateway. For cost per feature and per customer with margin, LLMtrack.
Can I use a gateway and a cost tracker together?
Yes. LLMtrack reads traces from LiteLLM, OpenRouter and Cloudflare AI Gateway with one URL, so the gateway keeps routing and LLMtrack prices each request. See all integrations.
Do I need a tool at all?
For one model and one feature, logging token counts to your own database is probably enough. A tool starts to pay off when you have several models or features, or need prices kept current as providers change them.
Which of these store prompts?
Observability platforms and gateways are built to record requests, which usually includes prompts and responses. LLMtrack only receives token counts, the model, a feature name and an optional customer ID.
Free plan with every feature. One async call, no proxy, no card.