How much does Portkey cost?
Portkey is free up to 10,000 logged requests a month and $49 a month on Production for 100,000, then $9 per extra 100,000; Enterprise is priced on request. Those plans cover the gateway and its logs only: tokens are billed separately by the providers whose keys you connect. Standard Compute is a gateway that includes the compute, in plans from $19 a month.
Portkey's plans
| Plan | Price | Logged requests | Log retention | Budgets per key |
|---|---|---|---|---|
| Developer | Free | 10,000 a month; requests past it still run, unlogged | 3 days | No |
| Production | $49 a month | 100,000, then $9 per 100,000 up to 3 million | 30 days | Select customers |
| Enterprise | Custom | 10 million and up | Custom | Yes |
| Open source | Free, self-hosted | No limit | Yours | Build it yourself |
From portkey.ai/pricing and Portkey's budget docs, October 13, 2026. Portkey calls Developer "not suitable for production workloads". The open-source gateway's last commit was May 25, 2026.
What a coding team pays in logs
The log fee stays small next to the token bill. Portkey can cap that bill, but it doesn't lower it: routing is conditions, fallbacks and load-balancing weights you write, so every request goes to the model your config names.
| Team | Logged requests a month | Portkey plan | Plus provider tokens |
|---|---|---|---|
| 5 developers, typical days | 9,400 | Developer, $0 | About $353 on Sonnet 5.5 |
| 20 developers, typical days | 37,600 | Production, $49 | About $1,412 |
| 20 developers, busy days | 196,400 | Production, $58 | About $11,588 |
| 50 developers, busy days | 491,000 | Production, $85 | About $28,970 |
One logged request per agent request: 94 a day for a typical developer and 491 on a busy day, over 20 working days. Provider tokens at Anthropic's list price for Claude Sonnet 5.5. Typical and busy are the median and 90th-percentile coding-agent day in our Agent Burn Index (September 13 to October 12, 2026; Claude Code and Codex not in the sample), over 20 working days.
Budgets and the Palo Alto change
- Budget limits per API key and per workspace are "available on Enterprise plan and select Pro customers". At the limit, further requests are blocked until the weekly or monthly reset.
- A budget on a provider integration expires the key when reached; Portkey's docs disagree on whether that budget can be edited afterwards.
- Palo Alto Networks closed its acquisition of Portkey on May 29, 2026 and made it generally available as Prisma AIRS AI Gateway on July 16. It has published no Prisma AIRS price; portkey.ai still lists the plans above.
- SOC 2, ISO 27001, HIPAA, BAAs and VPC hosting are on Enterprise.
The catch
Portkey's price is the smallest line on the bill. A team picking Portkey still opens an account with each provider, pays their list prices, and decides which model every route uses.
Where Standard Compute fits
Standard Compute sits where Portkey sits, between your agents and the models, and answers the catch above by selling the compute too: one plan per key from $19 a month, and a router that moves routine steps to cheaper models so the token bill itself shrinks. When guardrails, a VPC data plane or HIPAA are part of a Palo Alto rollout, choose Portkey.
FAQ
Is Portkey free?
The Developer plan is free with 10,000 logged requests a month and 3-day log retention, and the open-source gateway is free to self-host. Model usage is never included: you pay your providers.
Does Portkey charge a markup on tokens?
No. Portkey doesn't sell tokens. You connect your own provider keys, the providers bill you, and Portkey's plans price logging, retention and features.
What happens past 10,000 logs on the free plan?
Requests keep working. Portkey's pricing page says only the logs beyond the limit are not recorded.
How much does Prisma AIRS AI Gateway cost?
Palo Alto Networks hasn't published a price; its page is contact-only. On October 13, 2026 portkey.ai still listed Developer (free), Production ($49 a month) and Enterprise (custom).
Does Portkey route to cheaper models automatically?
Not per request. Its docs describe conditional routing on metadata, parameters or URL path, plus fallbacks and load balancing, all configured by you.