Portkey alternatives after Palo Alto Networks: the plan and the tokens, priced

A Portkey alternative has to replace two bills. Portkey charges a plan priced by recorded requests, and your providers bill the tokens at their list price through the keys you give it. On 2026-10-22, 100,000 requests of 2,000 input and 500 output tokens on Claude Sonnet 5.5 come to $900.00 in tokens at Anthropic's list plus $49 for Portkey Production, $949.00 in all. The same month is $390.00 here, with no plan and no provider key.

Palo Alto Networks closed its acquisition of Portkey on 2026-05-29, and Portkey's pricing page now opens with a Prisma AIRS banner while it still lists its plans (read 2026-10-22). This page totals one month on Portkey and on six alternatives, dates every figure, and maps Portkey's gateway features to what replaces each. For the wider field, see the full list of gateways ranked by the fee on top.

Three rows bill nothing per token: Portkey, LiteLLM and Cloudflare on your own keys. Their floor is the provider's list price, and the plan or the server sits on top. OpenRouter and Requesty bill list plus a percentage, and Vercel bills list with card fees aside. UplinkSwitch is the only row in the table priced from its own catalog, under list on 58 of the 75 chat models with a list price.

The baseline in every row is the developer's own price, as published on Anthropic's pricing page for the Claude models. A gateway that passes that price through cannot cost less than it, whatever its plan.

Palo Alto Networks announced on 2026-05-29 that it had closed the acquisition of Portkey, which it calls a pioneer in AI gateways. The release makes Portkey the core AI gateway of Prisma AIRS, the company's AI security platform. On 2026-07-16 its blog announced Prisma AIRS AI Gateway as generally available, six weeks after the deal closed.

On 2026-10-22 the Portkey pricing page carries a banner that reads "Portkey is now PRISMA AIRS AI Gateway, generally available for all enterprises". The same page still lists the Developer, Production and Enterprise plans and the open-source gateway, at the prices in the table above. Nothing on the three pages ends the self-serve plans or changes their prices.

What changed is the owner, and the reader both Palo Alto Networks pages write for. The release calls the gateway "a mission-critical control plane for the enterprise". If your company already buys Prisma AIRS, staying is the short path. If you chose Portkey for the $49 plan over keys you already pay for, the next section prices that month elsewhere.

The month is a production app's traffic. Every request sends 2,000 input tokens and gets 500 back, a chat turn with some context and nothing cached. At 100,000 requests that is 200 million input and 50 million output tokens. At 1,000,000 requests it is ten times that.

Portkey Production includes 100,000 recorded requests for $49 and adds $9 for each further 100,000, up to 3,000,000. The plan is therefore $49 at the first volume and $130 at the second. The token column is the developer's list price, which reaches you through your own provider keys.

On Claude Sonnet 5.5 the plan is 5.2% of the Portkey total at 100,000 requests and 1.4% at 1,000,000. The rest is tokens at Anthropic's list. Swapping Portkey for OpenRouter or Requesty swaps a flat plan for a percentage. For the smaller month the three totals are $949.50 on OpenRouter, $945.00 on Requesty and $949.00 on Portkey Production.

The plan is the small number in every row above, and the rate per token is the large one. A gateway that only changes the fee moves the total by about the size of the plan; a lower rate per token moves the rest. Here the Sonnet month is $390.00 at 100,000 requests and $3,900.00 at 1,000,000, against $9,130.00 on Portkey Production at the larger volume. Every figure in the UplinkSwitch column comes from the live catalog, and the price index lists every chat model the same way, next to its list price.

Portkey sells a gateway, and a gateway is a list of jobs. Before you move, check which of them your app uses. Each row below names the Portkey feature with its plan, then every alternative whose page states an equivalent.

Two Portkey features do not appear on every row: guardrails and the prompt library. If your app depends on Portkey's guardrails to block or rewrite requests, look at Requesty's Enterprise plan, a self-hosted LiteLLM, or OpenRouter, whose docs list a Guardrails feature and API (read 2026-10-22). The rest of the list, from fallbacks to budgets, has a version on most rows.

The second question is who holds the provider keys. Portkey, LiteLLM and Cloudflare on your own keys leave you with an account at each developer, each with its own invoice, limits and key rotation. OpenRouter, Requesty and UplinkSwitch sell the tokens themselves, so one key and one balance replace them. The hosted LiteLLM alternative page works through what that changes for a team that rotates keys today.

• Guardrails in the request path: deterministic checks on the free plan, and LLM and partner guardrails on Production.
• A prompt library with templates, versions and variables, served by the same gateway as the requests.
• Simple and semantic caching on Production, which can serve a repeated request from the cache instead of sending it to the model again.
• An open-source gateway you host yourself with no request limit, so the same setup runs on a laptop and inside a VPC.
• Compliance in writing on Enterprise: SOC2 Type 2, GDPR, HIPAA and BAAs, with private cloud and VPC hosting.
• Your own provider contracts. Portkey bills nothing per token, so a negotiated rate with a developer or a cloud stays yours.
• Security teams that run Prisma AIRS get the gateway inside a platform they already buy (read 2026-10-22).

• No provider keys. One sk-uplink- key and one balance reach 271 chat models and 96 image, video, speech and 3D models, so there is no developer account to open, fund and rotate.
• Tokens under list: 58 of the 75 chat models with a list price sit under it on input and output, live from the catalog. On the month above that is $390.00 against $949.00.
• No plan and no log quota. Credit is prepaid from $10, with no subscription and no fee on the top-up, and each top-up locks your rates for 30 days.
• Spend control per key: a monthly limit, budget autopilot that answers with a cheaper model you name as the key nears that limit, and per-key request and concurrency limits.
• Records you can check: a signed receipt on every non-streamed answer and a signed monthly statement, both verified at statements/verify without an account.
• A daily real-model check that sends every model the same prompt and publishes the verdict (236 of 273 stable at the last run).
• Usage you can take with you: every request on the usage page with its tokens and charge, exported as CSV, with timing traces and a replay of any logged request on another model.
• Team members by role (owner, billing or member) on one account and one balance.

If your app reaches Portkey through the OpenAI SDK, switching keeps the SDK and changes two values: the base URL and the key. Remove any gateway-specific headers or config your Portkey setup added; nothing replaces them on this side. Copy model ids from the Models page; they drop the vendor prefix.

export OPENAI_BASE_URL="https://api.uplinkswitch.com/v1"
export OPENAI_API_KEY="sk-uplink-live-..."

from openai import OpenAI

# Same SDK, two values changed. Drop any headers or config your Portkey setup added.
client = OpenAI(base_url="https://api.uplinkswitch.com/v1", api_key="sk-uplink-live-...")

reply = client.chat.completions.create(
    model="claude-sonnet-5-5",  # ids from the Models page, no vendor prefix
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)

Claude Code and the Anthropic SDK use the Anthropic format on the Claude models: set ANTHROPIC_BASE_URL to the same base URL and ANTHROPIC_AUTH_TOKEN to the key. The SDK setup guide covers the OpenAI, Anthropic and Gemini SDKs one by one, at the same price in every format.

Move one service at a time. Point a staging key at the new base URL, send the requests your app sends, and compare the answers and the bill on the usage page before you move production. A key per service also keeps each service's spend apart on the usage page.

Keeping a gateway in your own network is a separate choice from where you buy tokens. The OpenRouter vs LiteLLM comparison works out when self-hosting pays, and the Requesty alternatives page prices a coding month for teams that route Roo Code and Cline through a hosted router. What this key stores per request is on the no-logs page.

Is Portkey free?

Partly. The Developer plan is free forever with 10,000 recorded logs a month; past that limit requests still go through but are not logged, and Portkey says the plan is not suitable for production. The open-source gateway is free to host yourself with no request limit. Neither makes the tokens free: your providers bill them at their list price through your own keys (portkey.ai/pricing, read 2026-10-22).

Who owns Portkey now?

Palo Alto Networks. Its press release of 2026-05-29 announced that it had closed the acquisition of Portkey and would make it the core AI gateway for Prisma AIRS. Its blog announced Prisma AIRS AI Gateway as generally available on 2026-07-16, six weeks after closing.

Is Portkey still sold on its own?

Yes, as of 2026-10-22. Portkey's pricing page lists Developer (free), Production ($49 a month) and Enterprise (custom) next to a banner that reads "Portkey is now PRISMA AIRS AI Gateway, generally available for all enterprises". Neither Palo Alto Networks page we read says the self-serve plans will end.

Which open-source gateway replaces Portkey?

LiteLLM is the open-source gateway named in Google's AI Overview and in the Merge, Braintrust and Concentrate lists of Portkey alternatives on its first page (2026-10-22): free to host yourself, with virtual keys, budgets, rate limits, fallbacks and logging (litellm.ai/pricing, read 2026-10-22). Portkey's own gateway is open source too and runs on your own machines with no request limit. Both route your own provider keys, so the tokens stay at list price.

Is there a Portkey alternative without provider keys?

Yes: a gateway that sells the tokens itself. OpenRouter bills list plus 5.5% when you buy credit and Requesty bills list plus 5% on model cost (both read 2026-10-22). UplinkSwitch sells from prepaid credit at its catalog rate, under list on 58 of the 75 chat models with a list price, with no fee on top-ups from $10.

Does Portkey charge per token?

No. Portkey prices its plans by recorded requests: 10,000 a month free on Developer, 100,000 for $49 a month on Production plus $9 per further 100,000 up to 3 million, and custom volumes on Enterprise. Your providers bill the tokens at their own rates (read 2026-10-22).

Portkey, Prisma AIRS and Palo Alto Networks, LiteLLM, OpenRouter, Requesty, Cloudflare and Vercel are trademarks of their owners. This page quotes Palo Alto Networks' press release of 2026-05-29 and blog post of 2026-07-16, and the public pricing pages of Portkey, LiteLLM, OpenRouter and Requesty as read on 2026-10-22, and of Cloudflare and Vercel as read on 2026-10-19. The one-to-one page UplinkSwitch vs OpenRouter has the full OpenRouter terms.

Our figures come from the live catalog at GET https://api.uplinkswitch.com/v1/catalog, the daily check at GET https://api.uplinkswitch.com/v1/proof and the terms on /terms. Get a key: prepaid from $10, no subscription, 30-day price lock, in the console.