OpenRouter vs Portkey vs TokenLab: Routing, Observability, and Cost

OpenRouter, Portkey, and TokenLab sit at different layers of the AI stack. OpenRouter is a runtime routing gateway that forwards requests to model providers under one API. Portkey is a runtime gateway plus an observability and governance layer (logging, guardrails, caching). TokenLab is a pricing and capability comparison resource, not a runtime gateway. If you're trying to answer "how much will this cost me," the honest answer is: TokenLab's live pricing evidence shows model cost per 1M tokens ranges from roughly $0.27 to $60 depending on the model you route to, but OpenRouter's routing markup and Portkey's pricing tiers are not published in a form this evidence set can verify. You need to check those two vendors' own pricing pages before budgeting a production workload.

This article gives you the concrete numbers TokenLab can verify, states plainly what it cannot verify, and gives you a checklist for the integration questions (endpoint, auth, error handling) that block most teams from shipping against a new gateway.

Key Takeaways

Source Snapshot

Source	What it covers	Observed at
TokenLab live model/pricing evidence	Per-token pricing across Anthropic, OpenAI, Google, DeepSeek, and other providers in TokenLab's catalog	2026-07-09
OpenRouter official docs	Routing mechanics, fallback behavior, authentication	Not independently verified in this evidence set; confirm directly before integrating
Portkey official docs	Pricing tiers, logging retention, guardrail scope	Not independently verified in this evidence set; confirm directly before integrating

Because no OpenRouter or Portkey official pricing/API documentation was supplied as evidence for this refresh, every claim about their fee structure below is marked as unverified and paired with a verification step instead of a fabricated number.

What Each Tool Actually Does

OpenRouter is a routing layer. You send a request formatted for one API, and OpenRouter forwards it to the provider and model you specify, or to a fallback model in your list if the primary is unavailable. This is useful if you want to route between models such as Claude Sonnet 5, GPT-5.5, Gemini 3.5 Flash, or open-weight models like DeepSeek V4 Pro without rewriting client code per provider SDK. Confirm current supported models and fallback behavior on OpenRouter's own docs, since model catalogs change frequently.

Portkey layers request logging, semantic caching, guardrails (content filtering, PII redaction), and prompt management on top of multi-provider routing. The core value proposition is audit and governance tooling around requests, tracing a given output back to a user, cost center, or prompt version, not just routing itself.

TokenLab is not a runtime gateway. It is a comparison resource: current pricing and model capability data across providers, so you can decide before you integrate. If you're deciding between OpenRouter and Portkey, or between five model providers, TokenLab's compare page is built for that decision point, not for handling production traffic.

LLM Gateway Comparison: Feature Matrix

Capability	OpenRouter	Portkey	TokenLab	Verified in this evidence set
Unified multi-provider API	Yes (per general vendor description)	Yes (per general vendor description)	N/A, comparison tool only	No, confirm on vendor docs
Automatic failover routing	Yes (per general vendor description)	Yes (per general vendor description)	N/A	No, confirm on vendor docs
Request-level logging/tracing	Limited (per general vendor description)	Extensive (per general vendor description)	N/A	No, confirm on vendor docs
Guardrails / PII redaction	Limited (per general vendor description)	Yes (per general vendor description)	N/A	No, confirm on vendor docs
Published routing markup %	Not found in this evidence set	Not found in this evidence set	Not applicable, no markup	No
Cross-provider pricing comparison	No	No	Yes	Yes, TokenLab live evidence, observed 2026-07-09

This table is intentionally conservative. Every "Limited" or "Extensive" label reflects general vendor positioning, not a measured feature audit, and is not benchmarked in this evidence set. Treat it as a starting checklist for your own trial, not a final scorecard.

If you're deciding this week: compare current per-token model pricing on TokenLab's model pages before you pick a gateway, since your total bill depends far more on which model you route to than which gateway forwards the request.

TokenLab Model Pricing: What 1M Tokens Actually Costs

This table uses TokenLab's live pricing evidence directly. It shows the cost to process 1 million input tokens plus 1 million output tokens for text models currently in the catalog. This is model-level cost, not a gateway fee. It answers "how much will the model itself cost me at volume," which is the number that dominates total spend regardless of which gateway you route through.

Model	Input $/MTok	Output $/MTok	Cost for 1M in + 1M out	Source	Observed at
DeepSeek V4 Flash	$0.09	$0.18	$0.27	TokenLab live pricing evidence	2026-07-09
Qwen3.7 Plus	$0.32	$1.28	$1.60	TokenLab live pricing evidence	2026-07-09
MiniMax M3	$0.30	$1.20	$1.50	TokenLab live pricing evidence	2026-07-09
DeepSeek V4 Pro	$0.435	$0.87	$1.305	TokenLab live pricing evidence	2026-07-09
GLM-5.2	$0.93	$3.00	$3.93	TokenLab live pricing evidence	2026-07-09
Kimi K2.7 Code	$0.74	$3.50	$4.24	TokenLab live pricing evidence	2026-07-09
Gemini 3.5 Flash	$1.50	$9.00	$10.50	TokenLab live pricing evidence	2026-07-09
Claude Sonnet 5	$2.00	$10.00	$12.00	TokenLab live pricing evidence	2026-07-09
GPT-5.5 Batch/Flex	$2.50	$15.00	$17.50	TokenLab live pricing evidence	2026-07-09
Claude Opus 4.8	$5.00	$25.00	$30.00	TokenLab live pricing evidence	2026-07-09
GPT-5.5	$5.00	$30.00	$35.00	TokenLab live pricing evidence	2026-07-09
Claude Fable 5	$10.00	$50.00	$60.00	TokenLab live pricing evidence	2026-07-09

Two things follow from this table. First, the spread between the cheapest and most expensive model in the catalog is roughly 220x on a per-1M-token basis, which dwarfs any plausible gateway markup. Second, whatever OpenRouter or Portkey charges on top of these rates, you should ask each vendor for that number directly, since neither is documented in this evidence set. A gateway markup of even a few percent on a $60 flagship-model bill is a materially different dollar amount than the same markup on a $0.27 DeepSeek V4 Flash bill, so the model choice matters more than the gateway choice for total cost control.

Portkey vs OpenRouter: Routing and Observability Differences

Routing behavior is where OpenRouter and Portkey compete most directly.

OpenRouter's routing model lets you specify a primary model and a fallback list. If the primary provider errors or times out, the request retries against the next model in your list, per OpenRouter's general product description. The cost implication: fallback models often carry different pricing and different output quality (see the table above), so a failover event can silently change your per-request cost. Log which model actually served each request, not just which model you requested.

Portkey's routing supports similar fallback chains paired with a logging layer, so you can see when and why a fallback triggered and what it cost. If governance and audit trails matter more to your team than raw model breadth, this pairing is the practical differentiator over OpenRouter's simpler API surface.

Neither tool eliminates provider lock-in risk. Prompt formatting, function-calling schemas, and context-window behavior still differ across underlying models even when the gateway API is unified. Test your actual prompts against every model in your fallback chain before relying on automatic failover in production, since a fallback to a cheaper model can also mean a fallback to a model that fails your prompt format silently.

OpenRouter and Portkey API Integration: What to Verify Before You Call It

No official OpenRouter or Portkey endpoint, authentication header format, or request/response schema was supplied as evidence for this refresh. Publishing a guessed curl command or endpoint here would fail silently for anyone who copy-pasted it into production, so this section is a verification checklist instead of sample code. Run it against each vendor's current official API reference before you write a single line of integration code:

• Confirm the exact base endpoint URL and whether it is versioned (check for a version path segment that could change between releases).
• Confirm the authentication header name and format (API key header vs. bearer token vs. OAuth), and whether keys are scoped per project or per account.
• Confirm the exact model identifier string format for the models you intend to call (provider-prefixed slugs are common across gateways but the exact format varies).
• Confirm the request body schema, especially how fallback/model-list ordering is expressed if you plan to use automatic failover.
• Confirm rate limit headers and behavior on 429 responses, including whether the gateway enforces its own limit on top of the underlying provider's limit.
• Confirm behavior on 4xx errors (malformed request vs. auth failure vs. model-not-found) so your error handling branches correctly instead of retrying a request that will never succeed.
• Confirm behavior on 5xx and 503 responses from the upstream provider, and whether the gateway auto-retries or passes the error straight through to your client.
• Confirm timeout defaults and whether they are configurable per request.
• Build your own retry/fallback logic explicitly rather than assuming the gateway's default fallback chain matches your cost and quality tolerance, since a default chain optimized for uptime can route you to a model 100x more expensive than your primary without a visible warning.

This checklist applies equally to OpenRouter and Portkey. Do not go to production against either gateway until you've walked it against their current, official documentation.

Cost Structure: Where the Numbers Actually Diverge

Gateway pricing models differ in structure, not just rate, based on general vendor positioning:

• OpenRouter is generally described as passing through provider pricing with some routing markup on top. The exact markup percentage is not in this evidence set. Verify current pricing directly on OpenRouter's docs, since rates can vary per model and per provider.
• Portkey is generally described as pricing around request volume and feature tier (logging retention, guardrail usage), separate from the underlying model cost. The exact tier breakpoints and prices are not in this evidence set. Verify directly on Portkey's pricing page.
• TokenLab does not charge for routing since it is not a runtime gateway. It aggregates published model pricing (see the table above) so you can compare cost-per-token across providers before choosing a gateway.

The practical cost risk in a multi-model setup is not the base gateway fee, it is unmonitored fallback routing. If your fallback chain silently shifts requests to a pricier model during a primary provider outage, your bill can spike without an obvious cause in your own logs. Set up cost alerts on whichever gateway you pick, and re-check current published model rates against TokenLab's pricing pages periodically, since neither this article nor any static comparison stays current as providers update rates.

Limitations

• No official OpenRouter or Portkey pricing/API documentation was supplied as evidence for this refresh. Every claim about their markup percentage, pricing tiers, or exact API schema is marked unverified above and requires direct confirmation on each vendor's site.
• The feature comparison table (Yes/Limited/Extensive) reflects general vendor positioning, not a measured feature audit or benchmark. It is not benchmarked in this evidence set.
• TokenLab's pricing table reflects live evidence observed 2026-07-09. Model pricing can change without notice; re-verify before budgeting a production workload.
• This article does not include a worked OpenRouter or Portkey routing-markup calculation because no source number was provided. If you need that number for a budget decision, request it directly from each vendor's sales or docs team.
• No code samples for calling OpenRouter or Portkey's API are included, since exact endpoint, authentication, and payload schema evidence was not supplied. Use the verification checklist above instead of copying code from a third-party source.

Decision Checklist

• Do you need request-level audit logging for compliance? If yes, weight Portkey higher.
• Do you need the widest model catalog with minimal setup? If yes, weight OpenRouter higher.
• Have you tested your actual prompts against every model in your fallback chain, not just the primary?
• Have you gotten a written markup or fee quote from OpenRouter and Portkey directly, rather than assuming a percentage?
• Have you compared current per-token model pricing (see the table above) so you know the model-level cost floor before adding any gateway fee?
• Are you also evaluating models for specific workloads, like coding, image, or video generation, where capability differences matter more than routing? Check TokenLab's guides on best AI models for coding, best AI image models, and best AI video models before picking a default model per gateway.
• Have you set cost alerts to catch silent fallback-driven price increases?

Compare AI models by price using TokenLab's live pricing data before you integrate either gateway into production code.

FAQ

Is TokenLab a replacement for OpenRouter or Portkey? No. TokenLab is a comparison resource for evaluating pricing and capabilities across providers. OpenRouter and Portkey are runtime tools that route and log actual API traffic. Use TokenLab to compare model cost, then integrate whichever gateway fits your governance and routing requirements.

How much extra will I pay in gateway fees routing 1M tokens through OpenRouter vs Portkey? This evidence set does not include a published markup percentage or fee schedule for either vendor, so no specific number can be stated here without inventing one. Request a current fee quote directly from OpenRouter and Portkey. What can be stated with evidence is the model-level cost floor: routing 1M input plus 1M output tokens through DeepSeek V4 Flash costs $0.27 at TokenLab's observed rate, versus $60.00 through Claude Fable 5, before any gateway fee is added.

Where is OpenRouter's or Portkey's API endpoint and how do I authenticate? Not documented in this evidence set. Use the verification checklist above against each vendor's current official API reference before writing integration code. Do not rely on a third-party article for endpoint or auth details that can change without notice.

Which gateway is cheapest for high-volume production traffic? It depends on your model mix, retry rate, and whether you need Portkey's logging tier, none of which is quantified here for the gateway fee itself. Compare current published pricing directly on OpenRouter's and Portkey's own pricing pages, and cross-reference model-level rates against TokenLab's pricing comparison. Do not assume the gateway fee alone determines total cost; the model you route to (a 220x price spread in the table above) usually dominates.

If you're building a production AI product and need to compare current model rates before locking in a gateway, get started with TokenLab's live pricing data to check exact per-token costs before you write integration code.