Published Updated
In one or two sentences, before anything else.
Requesty is a London-based LLM gateway, run by Requesty Ltd and seed-funded by 20VC in 2025, that sells OpenAI- and Anthropic-compatible access to more than 600 models at list price plus a 5% top-up fee. It stands out for routing policies, spend controls and an EU endpoint in Frankfurt.
What it can do, one sentence each, and whether we actually tested it.
Per-token prices in its model list match the makers' list prices, and a 5% service fee is added to every top-up: 100 USD of credit costs 105 USD.
Per Requesty's pricing page and billing docs; no minimum fee or minimum top-up is stated.
Fallback chains with 0 to 10 retries per model and exponential backoff, weighted load balancing and latency routing, called as model="policy/name".
Per Requesty's fallback-policy docs; routing is set in the dashboard rather than per request.
Spend limits per key, user or group, approved-model lists, allowed-IP lists, and PII and secrets guardrails.
Per Requesty's docs; guardrail scanners carry a separate per-token charge on standard plans.
router.eu.requesty.ai keeps Requesty's own processing in Frankfurt; pairing it with an EU-region model such as bedrock/claude-sonnet-5-5@eu-central-1 keeps inference in the EU too.
Per Requesty's EU routing docs. A non-EU model on the EU endpoint still sends inference outside the EU.
Admins register MCP servers, choose which tools to allow and manage their keys; Claude Code, Cursor and Roo Code connect through router.requesty.ai/v1/mcp, with usage analytics.
Per Requesty's MCP docs. It proxies servers you add; it is not an MCP server for Requesty's own data.
No card needed: free models only, at 200 requests a day, with routing, caching and analytics. Its model list showed 13 zero-priced models on October 21, 2026.
Per Requesty's pricing page; no free credits are mentioned there.
Adds Anthropic cache breakpoints for you; on by default for traffic from Cline, Roo Code, Open WebUI and some other clients.
Per Requesty's auto-caching docs; a DataCamp tutorial notes caching only takes effect on Anthropic and Gemini models.
The requestyai/skills repository holds 10 SKILL.md skills for its API, routing policies, guardrails, MCP gateway, model discovery and Claude Code and OpenCode usage analytics.
The repository, created April 30, 2026, is not linked from Requesty's docs.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Graded in
Public tests other people ran: what they gave it and how it did.
Case 1
Input
Calls to openai/gpt-5 and anthropic/claude-sonnet-4 through router.requesty.ai/v1, with streaming, a dashboard fallback chain and auto_cache on Anthropic.
What happened
The base-URL swap worked and the session cost about 0.10 USD. Failover was configured but not exercised under a real outage, caching applied only to Anthropic and Gemini, and latency was not measured.
Case 2
Input
ANTHROPIC_BASE_URL=https://router.requesty.ai and ANTHROPIC_AUTH_TOKEN set to a Requesty key in ~/.claude/settings.json.
What happened
Claude Code talks to Requesty's Anthropic Messages endpoint, can switch to other models, and its spend appears in Requesty's analytics. No public test has measured the latency this adds.
Case 3
Input
A chat request to https://router.eu.requesty.ai/v1 with the model bedrock/claude-sonnet-5-5@eu-central-1.
What happened
Requesty processes the request in Frankfurt and the model runs in an EU AWS region. Prompt and output logging is still on by default on self-serve plans unless you turn it off for the key.
What it does badly, with the evidence, and what we could not settle.
Weak
Prompts and outputs are logged by default.
Evidence: Requesty's security page says content logging is on for self-serve plans, encrypted, stored in the EU and kept up to 30 days; it can be switched off per API key, and organisation-wide zero retention is available on written request.
Its SOC 2 status is unclear.
Evidence: The security page (September 2026) calls the SOC 2 Type II programme "in progress", while older docs text and competitor comparisons describe Requesty as SOC 2 compliant.
Little independent evidence.
Evidence: The only hands-on public write-up is a DataCamp tutorial without measurements. Requesty's overhead benchmark (16 ms against 55 ms for OpenRouter) is its own and gives no method; BenchLM describes a paid audit but publishes no per-platform numbers.
A small balance limits how many requests can run at once.
Evidence: Requesty's docs set the organisation's in-flight limit at the balance times a tier multiplier (5, then 10 after 250 USD of lifetime spend, 15 after 1,000 USD): a 5 USD balance allows 25 concurrent requests.
No video models and few image models.
Evidence: Its public model list on October 21, 2026 held chat models plus image generation through /v1/images for a small number of models such as azure/openai/gpt-image-1, and no video models.
Not for teams that need a finished SOC 2 report today, independently measured performance, or image and video generation.
Evidence: The SOC 2 programme is "in progress" by Requesty's own account, no neutral test exists, and the model list has no video models.
Uncertain
Minimum top-up, refunds and credit expiry.
Why unsure: The pricing page and billing docs do not state them, and the terms page renders only a heading without JavaScript.
Whether bring-your-own-key traffic carries a fee.
Why unsure: Requesty's pages state no BYOK fee, but Kilo, a competitor, wrote in May 2026 that Requesty adds 5% on BYOK.
Real uptime.
Why unsure: Its status page showed 100% for the last 30 days and no incidents from August to October 2026; a 99.99% uptime SLA appears only in third-party comparisons, and no outside monitor was found.
Company size and track record.
Why unsure: Respan, a competitor, estimated about five staff in March 2026; Requesty pivoted to a gateway in 2025, and the three AWS Marketplace reviews describe its earlier analytics product.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Operator | Requesty Ltd, London; co-founders Thibault Jaigu (CEO) and Daniel Trugman (CTO); 2 million GBP seed led by 20VC, September 2025 | Source: soapbox.vcchecked |
| Price against the makers | Makers' list prices, e.g. Claude Opus 5.5 at 4 / 20 USD and GPT-5.4 at 2.50 / 15 USD per million input / output tokens | Anthropic and OpenAI list the same prices. Source: router.requesty.aichecked |
| Platform fee | 5% added to every top-up (100 USD of credit costs 105 USD); guardrail scanners billed separately on standard plans | Source: docs.requesty.aichecked |
| Minimum top up and refunds | Not stated in the pricing page or billing docs; the terms page shows only a heading without JavaScript | Source: requesty.aichecked |
| Payment methods | Cards charged in USD, GBP, EUR, JPY or CNY; iDEAL, Bancontact, EPS, Satispay, Billie, UPI, Naver Pay; invoicing by ACH or wire for enterprise; no PayPal or crypto mentioned | Source: docs.requesty.aichecked |
| Free tier | Free plan, no card: free models only, 200 requests a day | Source: requesty.aichecked |
| Models | 600+ claimed; 803 chat model entries under 40 provider prefixes in its public list, many of them the same model on different clouds; no video models | Source: router.requesty.aichecked |
| API formats | OpenAI Chat Completions and Responses, Anthropic Messages, embeddings, images, audio; tool calling, structured outputs, auto caching, web search | No batch endpoint is listed in its docs index. Source: docs.requesty.aichecked |
| MCP | MCP Gateway at router.requesty.ai/v1/mcp proxies MCP servers your admins register; no Requesty MCP server for its own docs or account found in its docs | Source: docs.requesty.aichecked |
| Agent Skills | requestyai/skills on GitHub: 10 SKILL.md skills, not linked from its docs; Anthropic's Skills API not documented | Source: github.comchecked |
| Coding agents | Guides for Claude Code, Claude Cowork, Codex, OpenCode, Cline, Roo Code, GitHub Copilot, OpenClaw and the Claude Agent SDK | Source: docs.requesty.aichecked |
| Privacy | Prompts and outputs logged by default on self-serve plans, up to 30 days in the EU, switchable per key; SOC 2 Type II "in progress" | Source: requesty.aichecked |
| Status page | status.requesty.ai: 100% uptime over the last 30 days for every inference endpoint; no incidents in August, September or October 2026 | Source: status.requesty.aichecked |
| Concurrency limit | Balance times a tier multiplier of 5, 10 or 15; no requests-per-minute cap | Source: docs.requesty.aichecked |
Answer first, then why, an example, the limits and what to do next.
Requesty costs slightly more than the official APIs: it charges the makers' list prices and adds 5% to every top-up. Claude Opus 5.5 is 4 / 20 USD per million tokens on both, so through Requesty it works out to 4.20 / 21 USD once the fee is counted.
Limits
Its model list showed Claude Sonnet 5.5 cache reads at 0.20 USD per million tokens, double Anthropic's 0.10 USD, on October 21, 2026.
Requesty's free plan covers free models only, at 200 requests a day, with no card required. The current pricing page mentions no free credits; a competitor's market map from March 2026 described an earlier 6 USD credit.
Requesty is a registered UK company with an EU processing option, but it logs prompts by default and its SOC 2 audit is still in progress by its own account. You can turn logging off per key and ask for organisation-wide zero retention in writing.
Limits
Its terms page could not be read without JavaScript, so refund and suspension rules are unconfirmed.
Requesty suits teams that want dashboard governance and EU processing, while OpenRouter has the larger catalog and finer per-request routing. Requesty adds 5% to top-ups with no stated minimum; OpenRouter adds 5.5% and lists 686 models including video, against Requesty's chat-focused list.
Why
OpenRouter keeps prompt logging off by default and limits EU processing to its Business and Enterprise plans; Requesty logs by default but offers an EU endpoint on every plan.
Requesty supports MCP through its MCP Gateway, which proxies MCP servers your organisation registers. Admins choose which tools to expose and manage keys, and Claude Code, Cursor and Roo Code connect to router.requesty.ai/v1/mcp.
Limits
Its docs show no Requesty-run MCP server for its own account data, and its API references do not document passing remote MCP servers to models.
Requesty documents Claude Code through its Anthropic-compatible endpoint. Set ANTHROPIC_BASE_URL to https://router.requesty.ai and ANTHROPIC_AUTH_TOKEN to a Requesty key; the same setup works for the Claude Agent SDK.
This page is built from Requesty's own site, pricing page, docs, security page, status page and public model list, the makers' price pages, a DataCamp tutorial, competitors' comparisons and Hacker News comments, all checked on October 21, 2026. Requesty's claims are labelled as its own. Independent evidence about Requesty is thin, and the page says where.
Requesty Ltd is a London company that started in 2023 as a customer-feedback analytics startup and pivoted to an LLM gateway in 2025. It raised a 2 million GBP seed round led by 20VC in September 2025, when it reported 25,000 developers (Soapbox); its docs now claim 70,000 developers and 90 billion tokens a day.
It sells one OpenAI-compatible base URL (router.requesty.ai/v1) and one Anthropic-compatible one (router.requesty.ai) in front of what it calls 600+ models from 30+ providers. Its public model list held 803 chat model entries under 40 provider prefixes on October 21, 2026. The prefixes are hosts, so one model often appears several times: Claude Sonnet 5.5 is listed under Anthropic, Bedrock and Vertex. There are no video models.
The pitch is control rather than catalog size: fallback chains, weighted load balancing and latency routing set in a dashboard, spend limits per key, user or group, approved-model lists, guardrails for personal data and secrets, analytics, and regional endpoints in the EU, US and Asia-Pacific.
Requesty's pricing page says you pay for what your application spends, plus 5%. The fee is added at top-up: a 100 USD top-up costs 105 USD (billing docs). Per-token prices in its model list mostly match the makers' list prices:
| Model, per million input / output tokens | Requesty list price | Maker's list price |
|---|---|---|
| Claude Opus 5.5 | 4 / 20 USD | 4 / 20 USD (Anthropic) |
| Claude Sonnet 5.5, cache read | 0.20 USD | 0.10 USD (Anthropic) |
| GPT-5.4 | 2.50 / 15 USD | 2.50 / 15 USD (OpenAI) |
| Gemini 3.8 Flash on Vertex | 0.75 / 3.75 USD | 0.75 / 3.75 USD through December 31, 2026 (Google) |
| DeepSeek V4.1 Flash | 0.30 / 1.20 USD from DeepSeek; 0.20 / 0.60 USD on DeepInfra | 0.30 / 1.20 USD at peak hours (DeepSeek) |
Add 5% to every row. Requesty states no minimum fee, unlike the 0.80 USD minimum in OpenRouter's 2025 fee post. The Sonnet 5.5 cache-read figure is double Anthropic's, and Requesty does not say whether that is what it charges. Your concurrency limit is your balance times a tier multiplier, so a 5 USD balance allows 25 requests in flight (docs).
No independent latency, uptime or output-quality test of Requesty has been published.
Public user discussion is sparse. On Hacker News one account recommended Requesty for pinning providers (September 2026) and as the main EU alternative to OpenRouter (August 2026); another listed it among European alternatives (June 2026). On GitHub, many 2026 issues ask agent tools to add Requesty as a provider. No Trustpilot or Product Hunt listing turned up, and the three reviews on its AWS Marketplace listing date from 2024 and describe the earlier analytics product. Respan, a competitor, estimated about five employees in March 2026.
router.requesty.ai/v1/mcp, with tool allow-lists, organisation-wide keys (per-user keys on Enterprise) and usage analytics; Requesty documents it for Claude Code, Cursor and Roo Code (setup). No Requesty-run MCP server for its own docs, usage or account appears in its docs, and its Responses and Messages references do not document passing remote MCP servers to models.llms.txt index.Requesty's security page says prompts and outputs are logged by default on self-serve plans, encrypted and kept up to 30 days in the EU; you can turn logging off per key and request organisation-wide zero retention in writing. The EU endpoint keeps Requesty's processing in Frankfurt, but inference only stays in the EU if you pick an EU-region model. The same page calls its SOC 2 Type II programme "in progress", while older docs text and competitor comparisons describe Requesty as SOC 2 compliant.
Cards can be charged in USD, GBP, EUR, JPY or CNY, alongside iDEAL, Bancontact, UPI and Naver Pay; PayPal and crypto are not mentioned. The terms render only a heading without JavaScript, so refunds, credit expiry and blocked regions could not be checked. The status page showed 100% uptime for 30 days and no incidents from August to October 2026.
Skip Requesty if you need a finished SOC 2 report now, independently measured performance, video generation or written refund terms before paying. Teams that pick it for EU processing should turn off logging per key and choose EU-region models.
Sources
Changelog
APIMODELS is a pay-as-you-go AI API relay that sells about 150 models through one key at its own prices, mostly below the makers' list prices, and spends the margin on backup channels: 98.62% of requests succeeded over 30 days, moderation rejections excluded. APIMODELS is our own product: it is made by the team behind GenGrades.
ZenMux is an LLM gateway, operated under its terms by NexaMind Singapore Pte. Ltd., that sells Claude, GPT, Gemini and many Chinese models through OpenAI, Anthropic and Vertex-style endpoints at the makers' list token prices. It adds automatic credits for bad or slow answers, which it calls insurance, and 20 to 200 USD monthly coding plans.
Vercel AI Gateway is Vercel's hosted API for about 480 text, image, video and speech models behind one key, billed from prepaid credits at the providers' list prices with no markup, including when you bring your own keys. It became generally available on August 21, 2025, and gives each Vercel team 5 USD of free credit a month.
Poe API is Quora's developer API for the bots in its Poe chat app, opened on July 31, 2025, with OpenAI-compatible endpoints and an Anthropic-compatible one for official Claude bots. Calls spend the same points as a Poe subscription, so there is no separate API bill; current Claude bots cost about 1% more than Anthropic's list price.
OpenRouter is a New York company's API relay that sells access to more than 680 models from over 100 hosting providers through one key, at each provider's list price plus a 5.5% fee on credit purchases. Stripe agreed to buy it on August 19, 2026; no closing had been announced by October 21, 2026.
OpenCode Zen, called OpenCode Console in the docs since October 2026, is the pay-as-you-go model gateway from Anomaly Innovations, the team behind the OpenCode coding agent. It sells about 80 coding models near the makers' list prices plus a 4.6% + 0.31 USD card fee; a Console breach that month may have exposed some users' saved provider keys.