UplinkSwitch vs OpenRouter: same models, the fee and the proof

Pick OpenRouter for the largest model list (470 entries, read 2026-10-19), per-request provider choice and a $25,000 a month bring-your-own-key allowance. Pick UplinkSwitch, an AI API gateway, for the same model ids under the developer's list price, prepaid from $10 with no fee on the credit, a 30-day price lock, a daily real-model check and signed receipts. The one number: OpenRouter bills at list plus 5.5% on the credit, so Claude Sonnet 5.5 is $2.00 in and $10.00 out per million tokens there; here it is $1.20 and $3.00 on 2026-10-22.

OpenRouter bills each model at the developer's list price (Anthropic, OpenAI and the others publish it) and adds its fee when you buy credit, so its per-token column below is the list column. Ours comes from the live catalog at build time.

The full version of this table, every chat model on the key, is the LLM API pricing comparison; the price index downloads it as JSON and CSV. OpenRouter's own fee tiers and free-model limits are explained on its pricing page, and OpenRouter pricing explained works the fee at each top-up size.

On OpenRouter, $100 of credit costs $105.50 on a card (5.5%, $0.80 minimum, not refunded) and buys list-price tokens. On UplinkSwitch, $100 costs $100 and buys tokens at the catalog rate. Two shapes of request show what that is worth:

On Claude Sonnet 5.5 a Claude Code turn is $0.027 at list and $0.0129 here, so $100 is 3,703 turns on OpenRouter (after a $105.50 charge) and 7,751 here. The Claude Code pricing post turns the same shape into days and months.

Today's verdict for the Claude models in the table: Claude Opus 5.5, same fingerprint since oct 13; Claude Sonnet 5.5, same fingerprint since oct 12; Claude Haiku 5.5, same fingerprint since oct 19. Each verdict is the live one from the proof page, not a typed claim.

• Model count: 470 entries on its list against 267 chat models here (both read 2026-10-19). A model only OpenRouter serves settles the question.
• Provider ordering per request, fallbacks you control, and sorting by price, throughput or latency.
• Free model variants, rate-limited, for trying an open-weight model at no cost.
• Bring your own key at volume: $25,000 a month of list-price inference with no fee on pay-as-you-go (read 2026-10-19).
• Integrations that ship with an OpenRouter login: Roo Code's successors, LibreChat, Open WebUI and OpenCode list it as a provider out of the box.

• The same ids under list: 58 of 75 listed chat models with a list price, on input and output, from the live catalog.
• No fee on the credit, from $10, and a 30-day price lock on each top-up.
• Proof: the daily real-model check per model, and the canary prompt as public JSON for anyone to replay.
• Signed records: a receipt on every non-streamed answer and a signed monthly statement, verified in the browser at statements/verify.
• Spend control per key: a monthly limit, and budget autopilot that steps the key down to a cheaper model near the limit instead of refusing.
• One price in every wire format: the Anthropic SDK and Claude Code pay the same Claude rate as the OpenAI SDK (SDK setup).
• The uncensored builds and 96 media models on the same balance.

Change the base URL and the key; model ids drop the vendor prefix. Any tool that takes an OpenAI-compatible base URL works, and the SDK guide covers the OpenAI, Anthropic and Gemini SDKs one by one.

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

# Claude Code: Anthropic format, Claude ids
export ANTHROPIC_BASE_URL="https://api.uplinkswitch.com/v1"
export ANTHROPIC_AUTH_TOKEN="sk-uplink-live-..."

# Cursor: Settings > Models > OpenAI API Key, Override OpenAI Base URL = https://api.uplinkswitch.com/v1
# Cline and Zoo Code: API Provider "OpenAI Compatible", Base URL https://api.uplinkswitch.com/v1, one model id
# OpenCode: provider block with "npm": "@ai-sdk/openai-compatible", "baseURL": "https://api.uplinkswitch.com/v1"
# SillyTavern and Janitor AI: custom OpenAI-compatible source, https://api.uplinkswitch.com/v1

Guides: Claude Code, Cursor, Cline, OpenCode, SillyTavern and Janitor AI. For the whole field, OpenRouter and eight alternatives ranked by the fee on top, read OpenRouter alternatives compared by fee.

What does OpenRouter charge on top of the model price?

Nothing per token, by its own FAQ, and a platform fee when you buy credit: 5.5% on the Standard plan and 8% on Business (openrouter.ai/pricing, read 2026-10-19), with a $0.80 minimum on card purchases and 5% on crypto (its FAQ, read 2026-10-12). A $100 credit purchase is a $105.50 charge, and the fee is not refunded.

Is OpenRouter cheaper than the Claude API direct?

No. OpenRouter bills Claude Sonnet 5.5 at Anthropic's list price, $2.00 in and $10.00 out per million tokens, and adds 5.5% to the credit, so it costs slightly more than Anthropic direct. UplinkSwitch bills the same model at $1.20 and $3.00 with no fee on the credit.

Does OpenRouter have rate limits?

Its pricing page says it sets no plan-based limit on paid models and that a limit you hit normally comes from the provider; free models are capped at 20 requests a minute and 50 a day, rising to 1,000 a day once the account has bought $10 of credits (read 2026-10-19). On UplinkSwitch every key carries a requests-per-minute and a concurrency limit you set in the console, and a 429 carries retry-after.

Do OpenRouter credits expire?

OpenRouter reserves the right to expire unused credits one year after purchase and refunds unused credits only within 24 hours, never for crypto (its FAQ, read 2026-10-19). UplinkSwitch's terms set no expiry on prepaid credit; top-ups are not refundable.

Which model ids change when I switch?

The vendor prefix goes and the version uses dashes where the developer does: anthropic/claude-sonnet-5.5 becomes claude-sonnet-5-5, z-ai/glm-5.3 becomes glm-5.3. The table above maps the ten most used ids; the Models page lists every id the key can call.

Can I keep per-request provider choice?

No. OpenRouter lets a request order providers, turn off fallbacks or sort by price, throughput or latency. UplinkSwitch runs each model on one managed route at one price, checks the model daily and signs every answer; the route is never named.

Is the model really the one I asked for?

On UplinkSwitch the daily real-model check sends each model the same fixed prompt and compares the reported input token count, a fingerprint set by the developer's tokenizer; the verdict per model is public at /proof and the canary prompt is public JSON at https://api.uplinkswitch.com/v1/proof, so anyone can replay it against the developer's own API. OpenRouter's FAQ and pricing page state no such check (read 2026-10-19).

OpenRouter is a trademark of its owner; this page quotes its pricing page, FAQ and model list as read on 2026-10-19, and its provider-routing docs as read on 2026-10-12. Our figures come from GET https://api.uplinkswitch.com/v1/catalog, GET https://api.uplinkswitch.com/v1/proof and /terms. Get a key: prepaid from $10, no subscription, 30-day price lock, in the console.