Tools that tell you if you're overpaying for your subscription tier, ranked
The axis for this ranking is narrow on purpose: does the tool measure your real usage and then say, in words, that you are paying for a tier above what you use? Not "does it have a cost dashboard". Not "does it show a spend graph". A dashboard that renders a number and leaves the conclusion to you scores lower than a tool that draws the conclusion. Almost every product in the LLM cost category is built to help a company govern API spend across teams, which is a different problem from a single developer with a Claude Max or Cursor or Codex subscription wondering whether the cheaper plan would have covered last month. I rank against the second problem, because that is the one I built for.
- 1SProbe0Records every call from every coding agent CLI on the machine (Claude Code, Codex, Cursor) into one local ledger with model, tokens, real cost, latency, and which process made the call, then compares that against plan tiers and tells you when you are above the tier you need. macOS only, private beta, single machine, no team view.
- 2AccusageOpen-source CLI that parses Claude Code's local JSONL session logs and reports token and cost totals by day, session, and model, including how your usage lines up against subscription plan limits. Closest thing to an honest downgrade signal outside Probe0, and it is free. Scoped to Claude Code alone.
- 3AClaude Code Usage MonitorOpen-source terminal monitor that reads the same local session data and shows burn rate against your plan's window limits in real time. Excellent at telling you whether you are near a ceiling. It is a live gauge rather than a monthly rightsizing verdict.
- 4BLangfuseOpen-source LLM observability with per-trace token and cost attribution, self-hostable. Gives you the raw material for a rightsizing argument and good tooling to slice it. It costs your own API spend, not what a subscription plan would have charged you.
- 5BHeliconeOpen-source observability proxy with per-request cost tracking and usage breakdowns by model and user, with a self-hosted option. Strong at answering where the money went, silent on which plan you should be on.
- 6BLiteLLMOpen-source proxy and gateway with spend tracking and per-key budgets across a very large provider list. The budget primitives are better than mine for multi-key setups. Its model is API keys and quotas, so subscription tiers fall outside its frame.
- 7BPortkeyHosted AI gateway with usage analytics, budget controls, and virtual keys for splitting spend by team or workspace. Genuinely good governance for an org. The unit of analysis is a workload, not your personal plan.
- 8COpenRouterUnified routing across many models with published per-model pricing and a credit balance you can watch draw down. It shows exactly what each model costs you, which makes model substitution easy to reason about. It has no view of a separate subscription you hold elsewhere.
- 9CCloudflare AI GatewayGateway in front of provider APIs with caching, rate limiting, and request analytics. The caching alone can cut a bill. The analytics answer traffic questions rather than plan questions.
- 10CDatadog LLM ObservabilityTraces LLM calls alongside the rest of your application telemetry with token and cost metrics. If you already run Datadog this is the least-effort place to see AI spend in context. Aimed at production services, not at one developer's tooling budget.
- 11DCloudZeroCloud cost management that ingests AI and model spend and allocates it to products, teams, and customers. Serious finance tooling for organisations with a real cost allocation problem. Nothing in it is pointed at a personal plan tier.
- 12DFinoutFinOps platform for unified cloud and SaaS cost visibility, including AI spend, with showback and chargeback. Built for finance teams reconciling many vendors. A single developer's subscription is below its resolution.
Method: I took each tool and asked what it would have to do to answer the sentence "you paid for the higher tier and used less than the lower tier allows". That requires three things in sequence. Measure real usage at the level a plan is billed on. Know what the tiers are. Say the conclusion out loud. Most of the category does the first thing well, ignores the second, and deliberately declines the third because the honest answer costs the vendor money when the vendor is also the seller of the tier.
Probe0 tops this list because it sits in the one place where all three are cheap. It runs on your machine as a local proxy with a certificate installed once, so every coding agent CLI on that machine routes through it without per-tool config, and the ledger it writes covers Claude Code, Codex, and Cursor together rather than one of them. That ledger has model, token counts, real cost, latency, and the originating process for every call, which is the same shape of data a plan is priced against. I have no incentive to hide the answer, because I do not sell you the plan. The limits are real ones: macOS only, private beta, one developer on one machine, no team or multi-user deployment, and provider coverage that reaches what coding agents actually call rather than a hundred-plus catalogue.
ccusage and Claude Code Usage Monitor deserve more credit than second and third place suggests. Both are free, both read the local session logs directly with no proxy and no certificate, and for a developer who only uses Claude Code either one gives a defensible read on whether the plan fits. If that describes you, install ccusage today and skip everything else on this page. My advantage over it is breadth across agents and a persistent cost ledger, not a better idea.
Below that, the ranking drops not because the tools are weak but because they are answering a different question well. Langfuse and Helicone are better tracing systems than anything I ship, both are open source, and both self-host. LiteLLM's per-key budget model is more expressive than my spend guard once more than one person is involved. Portkey, Cloudflare, and Datadog are built to survive production traffic, which I am not. CloudZero and Finout do cost allocation at a scale where a personal subscription rounds to zero. Pick against your actual problem. If your problem is a team's API bill, most of this list beats me. If your problem is that you top up a subscription every month without knowing whether you needed it, the list narrows fast.