What is the difference between Portkey and PromptUnit?

Portkey is a gateway you configure: routing rules, fallback between providers, load balancing, request logs and prompt management. PromptUnit checks every request and uses a cheaper model when one is just as good for it, with no rules to write, and charges 20% of what it saves.

Does Portkey route requests to cheaper models automatically?

Portkey routes by the rules you set: conditions on request data, weights for load balancing and an order for fallbacks. Its documentation describes no automatic per-request choice of a cheaper model (checked October 15, 2026).

How much does Portkey cost?

Portkey has a free Developer plan with 10,000 logs a month, a Production plan at $49 a month for 100,000, and custom enterprise pricing; its open-source gateway is free to self-host (checked October 15, 2026). PromptUnit has no monthly fee: it takes 20% of what it saves, and nothing if it saves nothing.

Does PromptUnit switch providers when one is down?

No. If a cheaper model PromptUnit chose fails, it retries on the model you asked for, but it does not reroute a request to another provider during an outage. For provider failover, use a gateway such as Portkey or build your own.

PromptUnit vs Portkey: Same Category, Different Priorities

Portkey and PromptUnit are the most similar tools in this comparison series. Both are LLM gateways. Both sit between your application and providers. Both route requests between models; only Portkey switches providers when one fails, while PromptUnit retries on the model you asked for. Their design priorities are different in ways that matter for real purchase decisions.

Portkey is built for control, reliability, and observability across LLM providers.

PromptUnit is built for cost reduction, specifically, cutting LLM inference spend automatically with quality validation.

What Portkey Actually Is

Portkey is a managed LLM gateway with a wide feature set: routing, fallback, load balancing, observability, prompt management, and caching. It positions itself as the control plane for LLM infrastructure.

Routing in Portkey is configuration-based. You define rules, "if provider A fails, route to provider B" or "send 20% of traffic to this model and 80% to that one." The rules are powerful and flexible. The trade-off is that you define them manually; Portkey does not automatically discover which requests could route to cheaper models.

What Portkey does well:

Comprehensive gateway features in one platform
Flexible rule-based routing and fallback configuration
Full request/response logging and observability
Prompt management and versioning
Load balancing across multiple providers and deployments
Guardrails for output validation
Enterprise SLA and compliance features

What Portkey does not do:

Automatically classify requests by task type to find cost reduction opportunities
Make routing decisions based on inferred complexity, you configure the rules
Show you the savings on each request, starting with the first one
Price itself based on what it saves you

What PromptUnit Actually Is

PromptUnit cuts LLM inference cost. Its engine checks every request by task type and complexity, then picks the cheapest model that clears your quality setting, automatically, with no routing rules to write.

The key mechanic is per-request checking. When a cheaper model is just as good for a request, PromptUnit uses it. When it isn't, the request goes to the model you asked for. Savings start with the first request. Pricing is 20% of verified savings, no fixed monthly cost.

What PromptUnit does well:

Automatic task classification and cost-optimizing routing
Quality-validated routing with configurable threshold
Savings from the first request, with every request shown on the dashboard
Cross-provider routing (OpenAI, Anthropic, Google, Groq, DeepSeek)
Savings attribution by feature, model, and task type
Pay-only-for-savings pricing

What PromptUnit does not do:

Manual rule-based routing configuration
Detailed request/response logging for debugging
Prompt versioning and A/B testing
Load balancing across your own model deployments
Enterprise compliance certifications (no SOC 2 report yet)

Comparison Table

Property	Portkey	PromptUnit
Primary purpose	LLM control plane	LLM cost optimization
Routing type	Rule-based (manual config)	Automatic (a check on every request)
Cost-optimizing routing	Manual rules only	Automatic, quality-validated
Per-request savings view	No	Yes, from the first request
Request logging	Full capture	Cost and metadata
Prompt management	Yes	No
Fallback	Yes (configured)	Retries on your requested model if a cheaper one fails
Self-hosted	Yes (open-source)	No (managed SaaS)
Pricing	Free plan (10k logs a month), Production $49 a month, enterprise plans; the open-source gateway is free to self-host (checked October 15, 2026)	20% of verified savings
Setup complexity	Medium (rule configuration)	Low (one address to paste)

Which to Choose

Choose Portkey if:

You need a full-featured LLM control plane with manual routing control
Request/response logging and debugging workflows are important
You want to define your own routing rules precisely
Self-hosted deployment is a requirement
You need enterprise compliance features now
Prompt management and versioning are part of your workflow

Choose PromptUnit if:

Your primary goal is reducing LLM inference costs automatically
You do not want to manually write and maintain routing rules
You want to see what each request cost and saved, starting with the first one
Pay-for-results pricing fits better than a fixed monthly fee
A setup of one address to paste, with zero routing configuration, is the priority

The Core Difference

Portkey gives you control over your LLM infrastructure. You decide the rules. It executes them reliably.

PromptUnit gives you cost reduction on your LLM infrastructure. It decides the routing based on request-level analysis. You decide the quality floor.

If your problem is: "I want full control over how my LLM traffic routes across providers", Portkey.

If your problem is: "I want my LLM bill to go down without building routing logic myself", PromptUnit.

For teams with existing LLM infrastructure who want to layer cost optimization on top: PromptUnit takes one change to add: you paste your personal PromptUnit address where your app's AI address (base URL) is set. Your AI key stays in your app; PromptUnit never asks for it and does not store or log it. Savings start with the first request. For teams building complex multi-model architectures who need precise routing control: Portkey's rule engine is more flexible.

Try It Free

One address to paste. Savings from the first request. Pay only from savings.

Start free, no credit card required.