OpenRouter vs Portkey: How AI Coding Teams Should Choose an LLM Gateway

Forum topic · QianXun · 2026-06-22

Summary

OpenRouter published back-to-back comparison posts on June 19, 2026, contrasting its managed routing network with Portkey's AI control plane and the self-hosted LiteLLM proxy. OpenRouter routes requests through a single API to 300+ models across 70+ providers with automatic failover, zero markup on model pricing, a 5.5% platform fee, Zero Data Retention, and EU regional routing. Portkey, acquired by Palo Alto Networks in 2026, sits above user-held provider keys and adds governance: logging, tracing, prompt management, RBAC, budgets, guardrails with PII redaction, plus HIPAA, custom BAAs, VPC hosting, and a free open-source gateway (12K+ GitHub stars). OpenRouter's own decision threshold: if monthly model spend is at or above $3,600, self-hosted Portkey is recommended; below that, pay-as-you-go OpenRouter is cheaper. The two are not mutually exclusive—Portkey can treat OpenRouter as an upstream provider. This piece also covers the OpenRouter vs LiteLLM comparison and what the consolidation means for AI coding infrastructure.

Background

On June 19, 2026, OpenRouter published two comparison blog posts: vs Portkey and vs LiteLLM. Notably, Portkey was acquired by Palo Alto Networks in 2026 and folded into its AI security product line — a key signal of industry consolidation.

Source: https://openrouter.ai/blog/insights/openrouter-vs-portkey

Core Positioning Difference

OpenRouter: a managed routing network. Buy credits, route through one API to 70+ providers, with automatic failover — no need to hold your own provider keys.
Portkey: an AI control plane. Sits on top of your existing keys and adds governance, prompt management, guardrails, and observability.

Bluntly put: OpenRouter helps you call models; Portkey helps you manage calls.

Model Coverage

Portkey: 1,600+ LLMs via a unified API
OpenRouter: 300+ models across 70+ providers, including 20+ free models for evaluation

The numbers differ, but they measure different things:

Portkey's count is the breadth you can reach (provided you hold your own provider keys)
OpenRouter's count is the breadth it actively routes to (automatic failover, filtering by price/throughput/latency/data policies)

Observability and Governance

Portkey centralizes governance: built-in logging, tracing, dashboards (3-day free / 30-day production / enterprise custom); prompt templates and playground; RBAC; budgets and rate limits; guardrails including PII redaction.
OpenRouter stays light: Broadcast streams traces to platforms you already use (Datadog, Langfuse, LangSmith, Braintrust, OpenTelemetry, S3); Workspaces provide team organization and budgets.

If you already have a Datadog pipeline, OpenRouter won't reinvent the wheel; if you're building governance from scratch, Portkey works out of the box.

Compliance

Both: SOC 2 Type 2 + GDPR
OpenRouter adds: Zero Data Retention, EU regional routing (trust.openrouter.ai)
Portkey adds: HIPAA, custom BAAs, VPC hosting, data isolation

Rule of thumb: healthcare data → Portkey; EU data residency → OpenRouter.

Pricing Models

| Dimension | OpenRouter | Portkey | |---|---|---| | Model pricing | Zero markup | Bring your own provider key | | Platform fee | 5.5% (5% BYOK; first 1M requests free, $0.80 minimum) | Charged per log | | Free tier | First 1M requests/month | Developer tier 10K logs/month (not for production) | | Production tier | Usage-based | $49/month + 100K logs (+$9 per extra 100K) | | Enterprise tier | — | Custom (HIPAA, SSO, private deployment) | | Self-hosted | — | Free open-source gateway (12K+ GitHub stars) |

Decision Threshold (OpenRouter's own advice)

> Monthly model spend >= $3,600 (infrastructure $200/month) → choose self-hosted Portkey > Monthly model spend < $3,600 → choose OpenRouter pay-as-you-go

Not mutually exclusive: Portkey can use OpenRouter as an upstream provider (governance + routing in one stack); OpenRouter will later add Portkey as a Broadcast target (traces flowing the other way).

OpenRouter vs LiteLLM (companion post)

OpenRouter: hosted on Cloudflare's edge, 5.5% platform fee (first 1M requests free), automatic failover.
LiteLLM: self-hosted proxy (Docker/PostgreSQL/Redis); data never leaves your network; free and open source, but you bear infrastructure costs (production deployments run roughly hundreds of dollars per month).

LiteLLM suits compliance scenarios where "data must never leave the internal network."

Why This Matters

1. AI coding infrastructure is maturing: LLM gateways are clearly stratifying — routing network (OpenRouter) / control plane (Portkey) / self-hosted proxy (LiteLLM). Each layer has its own customers. 2. Consolidation signal: Portkey's acquisition by Palo Alto Networks shows the enterprise AI gateway market is on big companies' radar. More acquisitions are likely. 3. Pricing transparency is improving: 5.5% platform fee vs per-log billing vs free self-hosting — AI coding teams are starting to calculate costs differently. 4. Compliance needs are diverging: EU data residency / HIPAA / Zero Data Retention are becoming key selection criteria, no longer a one-size-fits-all "generic LLM" play. 5. Composable ecosystem: OpenRouter + Portkey is not either/or — they can be stacked. AI coding teams can start with a single gateway and layer on governance as they scale.

Risks and Open Questions

Portkey's independence post-acquisition: after Palo Alto's integration, will the open-source version stay active? Will pricing change?
Cumulative cost of OpenRouter's 5.5% fee: long-term, a self-hosted proxy (LiteLLM) may cost less — if your team has the ops capability.
The step-change in multi-provider governance costs: moving from a single provider (OpenAI/Anthropic) to multiple providers causes governance and observability costs to jump, not grow linearly.
Quality consistency in automatic failover: how do you handle behavioral differences when routing across providers? OpenRouter's Auto Router uses NotDiamond for model selection, but real-world behavior differences remain an open problem.