AICosts.ai vs Portkey: Read-only cost aggregator vs LLM gateway
Portkey is an LLM gateway — you send your LLM requests through their proxy and get a unified API across 250+ models plus caching, routing, fallbacks, guardrails, and real-time observability. AICosts.ai is a read-only billing aggregator — you paste read-only API keys per provider and we pull invoice-accurate cost, usage, and breakdowns from 50+ providers with zero inference-path impact. Portkey is the tool you pick when you want to standardize and control how your app talks to LLMs. AICosts.ai is the tool you pick when you need finance-grade month-end cost visibility without owning any runtime component. Teams who need both run both.
AICosts.ai vs Portkey: feature-by-feature
| Feature | AICosts.ai | Portkey |
|---|---|---|
| Architecture | Read-only billing aggregator (pulls invoices + usage from vendor APIs) | Gateway/proxy — your app sends every LLM request through Portkey |
| Inference-path impact | Zero — never touches your request path | Gateway adds a hop; marketed as <50ms overhead, but it's on the hot path |
| Providers tracked (cost) | 50+ across LLMs + embeddings + vector DBs + automation (OpenAI, Anthropic, Bedrock, Vertex, Azure, Pinecone, Tavily, RunwayML, plus 40+ more) | 250+ LLM models via Portkey's unified API. Non-LLM services (vector DBs, automation tools, media generation) are not tracked. |
| Invoice reconciliation | Pulls actual invoice data — month-end numbers match the vendor bill | Costs estimated from published rate cards; drifts with prompt-cache, batch, committed-use, or negotiated pricing |
| Routing, caching, fallbacks, guardrails | Explicitly out of scope — we don't sit in your inference path by design | First-class: semantic caching, load balancing, automatic fallbacks, PII/toxicity guardrails, retries |
| Prompt management | Out of scope | Versioned prompts, partials, A/B testing across models |
| Setup | Paste a read-only API key per provider (no code changes) | Change your SDK's base_url to Portkey's gateway, or install the Portkey SDK |
| Self-hosted / VPC deployment | Cloud-hosted only today (self-host on roadmap) | Enterprise tier supports self-hosted / VPC deployment |
| Best fit | Finance, founders, AI FinOps — 'what did we actually spend across every vendor' | Platform/infra teams who want to standardize and control how their app talks to LLMs |
Where AICosts.ai is stronger
No proxy to own or operate
AICosts.ai never sits between your application and an LLM provider. There's nothing to fail, nothing to scale, nothing that can add latency to user-facing requests. Portkey's gateway is well-engineered, but it is still a runtime component you're responsible for — including their outage becoming your outage.
Invoice-accurate cost, not rate-card estimates
Portkey computes cost by multiplying token counts by public rate cards. When your vendor applies prompt-cache discounts, batch pricing, committed-use deals, or negotiated enterprise rates, that number drifts from the actual invoice. AICosts.ai pulls the invoice itself.
Covers every billed vendor, not just LLM calls
Your Pinecone bill, Tavily bill, RunwayML bill, ElevenLabs bill, Make/Zapier automation bill — none of those flow through Portkey because they're not LLM calls. AICosts.ai pulls all of them into one dashboard alongside your OpenAI and Anthropic spend.
Zero risk to adopt or remove
Because we're not in the request path, connecting or disconnecting AICosts.ai is a reversible, zero-stakes decision. Disconnecting an API key stops ingestion; nothing downstream is affected. Removing a gateway is a deploy.
Predictable pricing
AICosts.ai is flat monthly ($19.99 or $49.99), no per-request metering. Portkey at production scale scales with your request volume — on busy months the bill grows with it.
Where Portkey is a better fit
No tool is right for every problem. Here's when Portkey is the more honest pick.
No request routing, caching, fallbacks, or guardrails
If your goal is to standardize how your app talks to LLMs — caching similar requests, automatically falling back from GPT-4o to a cheaper model, enforcing PII redaction before a prompt leaves your infra — that's exactly what gateways like Portkey are for. AICosts.ai explicitly doesn't do any of that, by design.
No prompt management or versioning
Portkey ships versioned prompt templates with A/B testing across models. AICosts.ai has none of that.
No per-request logs or latency data
If you need to see the exact prompt, completion, and latency of a specific failing request, AICosts.ai can't help. Portkey's observability layer logs every request going through the gateway.
Daily granularity, not real-time
Billing APIs refresh hourly-to-daily. If you need sub-minute cost signal during a deploy, Portkey's real-time gateway metrics are closer. We give you end-of-day invoice accuracy.
Pricing at a glance
AICosts.ai
Starter $19.99/mo (up to 3 providers, 30-day retention). Professional $49.99/mo (all 50+ providers, 90-day retention, billing-file upload + parse). Enterprise is custom. 7-day free trial on Starter and Professional. Flat monthly, not metered by request.
Portkey
Portkey Developer tier is free with a monthly request cap. Production tier around $99/mo plus request-volume metering. Enterprise custom with SSO, VPC self-host, SLA. Priced by request volume through the gateway + feature tier.
Verdict
Pick AICosts.ai when your problem is 'we have 8 vendor invoices and no unified month-end number,' when invoice accuracy matters more than real-time per-request signal, when running a gateway in your hot path is a non-starter (regulated industry, compliance-sensitive path), or when you want predictable flat pricing. Pick Portkey when you want a unified API across 250+ LLMs, when caching/routing/fallbacks/guardrails are part of your architecture plan, or when prompt management and per-request observability are core to your workflow. Teams who need both run both: Portkey as the runtime gateway, AICosts.ai as the finance-grade cost layer that also covers non-LLM vendors.
Frequently Asked Questions
Can I use Portkey and AICosts.ai together?+
Yes, and it's a clean pairing for teams who need both control and finance-grade reporting. Portkey handles routing, caching, fallbacks, guardrails, and per-request observability. AICosts.ai handles invoice-accurate cost across LLM vendors and every non-LLM vendor (Pinecone, Tavily, RunwayML, etc.). The two surfaces don't overlap.
Why not just use Portkey's cost dashboard and skip a second tool?+
Two reasons: (1) Portkey's cost number is computed from public rate cards — once your vendor applies prompt-cache savings, batch discounts, committed-use pricing, or negotiated enterprise rates, the estimate drifts from what the invoice actually says. (2) Portkey only sees what goes through the gateway. Your AWS Bedrock calls using the AWS SDK directly, your Pinecone vector-db bill, your RunwayML and ElevenLabs media bills, your Make/Zapier automation bill — none of those are in Portkey.
Does AICosts.ai support gateway features like caching or fallbacks?+
No, and we won't — those are explicit non-goals. Our wedge is that we aren't in your inference path. Adding routing would mean becoming a proxy, which doubles our ops blast radius and puts our uptime in your request path. If you need gateway features, Portkey or a similar tool is the right choice.
Which tool is cheaper?+
They price on different axes. Portkey's free Developer tier covers experimentation; Production tier is ~$99/mo plus request-volume metering. AICosts.ai is $19.99 or $49.99 flat. For a team making 10M+ LLM requests per month through a gateway, Portkey's request-based pricing scales upward; for a team primarily needing multi-vendor cost visibility at modest volume, AICosts.ai is predictable.
Is a read-only billing aggregator actually reliable?+
Yes — it's how every mature FinOps platform (Cloudability, CloudZero, Vantage) works for AWS and GCP cost tracking. AICosts.ai applies the same pattern to the AI stack. The tradeoff is daily granularity instead of per-request, which is the right tradeoff for finance-grade cost-management use cases.
Try AICosts.ai
Read-only. 50+ providers. Free tier available.
Start tracking your AI spendFree tier available. Read-only ingestion. No changes to production.