LLM gateways
OpenRouter vs Requesty vs Portkey vs FullsendAI
OpenRouter and Requesty resell many providers behind one key. Portkey sits in front of your own provider keys and sells observability and governance; it is now part of Palo Alto Networks. FullsendAI is the one with a web chat and one-command setup for your coding agents on the same key. Unify, once a router, no longer offers one.
Which one fits
Choose OpenRouter if…
- You want the widest catalogue (456 models) and fine control over providers.
Choose Requesty if…
- You need data kept in the EU and per-request cost in every response.
- You're fine with 5% on model cost.
Choose Portkey if…
- You already have provider contracts and need logs, guardrails and governance across them, inside an enterprise security stack.
Choose FullsendAI if…
- You want one key and balance for the API, a web chat and your coding agents.
- You want no fee on usage or top-ups, and every model in the USA.
At a glance
Each cell links to its source and the date we checked it. Rows show only where we have a sourced fact for every gateway.
What it is
Models
OpenAI-compatible API
Anthropic Messages API
Claude, GPT and Gemini
Routing and failover
Prompt logging by default
- FullsendAI
- API prompts not stored unless you turn on 30-day body storage
- OpenRouter
- Prompts not logged by default; opt in for a 1% discountSource ·
- Requesty
- Self-serve: prompts and outputs logged for up to 30 days; can be turned off per keySource ·
- Portkey
- Logs requests by design: 3 days free, 30 days on ProductionSource ·
Audits and certifications
Web chat
Desktop coding agent
Own tuned models
| Feature | FullsendAI | OpenRouter | Requesty | Portkey |
|---|---|---|---|---|
| What it is | One API key, a web chat and one-command setup for coding agents | Router in front of many model providersSource · | Router and gateway in front of many providersSource · | Gateway and observability over your own provider keysSource · |
| Models | 4 models | 456 modelsSource · | 809 model routesSource · | Whatever your own provider keys reachSource · |
| OpenAI-compatible API | Yes (Chat Completions) | YesSource · | YesSource · | YesSource · |
| Anthropic Messages API | Yes; works with Claude Code | Yes; documents Claude Code setupSource · | Yes; documents Claude Code setupSource · | YesSource · |
| Claude, GPT and Gemini | Not yet | YesSource · | YesSource · | With your own keysSource · |
| Routing and failover | We pick the host, with automatic failover; no provider controls | Provider ordering, fallbacks and filters you controlSource · | Fallback and load-balancing policiesSource · | Fallbacks, load balancing, canary testsSource · |
| Prompt logging by default | API prompts not stored unless you turn on 30-day body storage | Prompts not logged by default; opt in for a 1% discountSource · | Self-serve: prompts and outputs logged for up to 30 days; can be turned off per keySource · | Logs requests by design: 3 days free, 30 days on ProductionSource · |
| Audits and certifications | No third-party audit report published yet | SOC 2 Type 2 (trust center)Source · | SOC 2 Type II “in progress”Source · | SOC 2 Type 2, ISO 27001, GDPR, HIPAASource · |
| Web chat | Yes | Yes (chatroom)Source · | NoSource · | NoSource · |
| Desktop coding agent | Not yet: one command sets up Claude Code, Codex and more | No desktop agent; setup guides for third-party agentsSource · | No desktop agent; setup guides for third-party agentsSource · | NoSource · |
| Own tuned models | Nef 4.0 and Sifu 3.5, tuned for security research and code | No; routers such as openrouter/autoSource · | NoSource · | NoSource · |
Fees
What each charges beyond the per-token price. Portkey bills per log; you pay model providers directly.
FullsendAI
- On usage
- None
- Card top-up
- None
- Crypto top-up
- None
OpenRouter
- On usage
- None stated
- Card top-up
- 5.5%, $0.80 minimum
- Crypto top-up
- 5%
- Source
- Source ·
Requesty
- On usage
- 5% on model cost
- Card top-up
- None stated
- Crypto top-up
- No crypto option listed
- Source
- Source ·
Portkey
- On usage
- Free to 10K logs a month; $49 a month for 100K, then $9 per 100K
- Card top-up
- None stated
- Crypto top-up
- No crypto option listed
- Source
- Source ·
| Provider | On usage | Card top-up | Crypto top-up | Source |
|---|---|---|---|---|
| FullsendAI | None | None | None | Our pricing |
| OpenRouter | None stated | 5.5%, $0.80 minimum | 5% | Source · Fees are charged when you buy credits, not per request. |
| Requesty | 5% on model cost | None stated | No crypto option listed | Source · “You pay for what your application spends, plus 5%.” |
| Portkey | Free to 10K logs a month; $49 a month for 100K, then $9 per 100K | None stated | No crypto option listed | Source · Billed per log, not per token: you pay providers directly. |
Changes in 2026
Who owns what now, sourced and dated.
Portkey. Acquired by Palo Alto Networks (closed May 2026); now sold as Prisma AIRS AI Gateway
Portkey's own pricing page reads “Portkey is now PRISMA AIRS AI Gateway”.
Unify. Router API no longer answers (HTTP 404 from its models endpoint)
Checked with a direct request on this date.
Prices, model for model
USD per million tokens, input / output, before each gateway's fees. Requesty's price is for the route named; for OpenRouter, the price most of its providers charge, with the lowest-priced endpoint beneath.
DeepSeek V4.1 FlashDeepSeek
GLM 5.3 FlashZ.ai
GPT OSS 120BOpenAI
GPT OSS 20BOpenAI
Kimi K2.6Moonshot AI
Qwen3.8 MaxAlibaba
DeepSeek V4 Pro 0813DeepSeek
Claude Opus 5.5Anthropic
Gemini 3.1 ProGoogle
| Model | FullsendAI | OpenRouter | Requesty |
|---|---|---|---|
| DeepSeek V4.1 FlashDeepSeek | Not offered yet | $0.30 / $1.20from $0.022 / $0.192 (fp8, Morph)deepseek/deepseek-v4.1-flashSource · | $0.30 / $1.20deepseek/deepseek-v4.1-flashSource · |
| GLM 5.3 FlashZ.ai | Not offered yet | $0.15 / $0.50from $0.0189 / $0.0625 (fp4, OpenInference)z-ai/glm-5.3-flashSource · | $0.15 / $0.50zai/glm-5.3-flashSource · |
| GPT OSS 120BOpenAI | Not offered yet | $0.15 / $0.60from $0.03 / $0.15 (fp4, Near AI)openai/gpt-oss-120bSource · | $0.15 / $0.60fireworks/gpt-oss-120bSource · |
| GPT OSS 20BOpenAI | Not offered yet | $0.07 / $0.15from $0.018 / $0.09 (fp8, Darkbloom)openai/gpt-oss-20bSource · | $0.07 / $0.30fireworks/gpt-oss-20bSource · |
| Kimi K2.6Moonshot AI | Not offered yet | $0.95 / $4.00from $0.39 / $1.80 (int4, Chutes)moonshotai/kimi-k2.6Source · | $0.95 / $4.00moonshot/kimi-k2.6Source · |
| Qwen3.8 MaxAlibaba | Not offered yet | $2.00 / $6.00qwen/qwen3.8-max-0902Source · | $2.00 / $6.00alibaba/qwen3.8-maxSource · |
| DeepSeek V4 Pro 0813DeepSeek | Not offered yet | $1.32 / $3.96from $0.675 / $1.20 (fp4, Sail Research)deepseek/deepseek-v4-pro-0813Source · | $1.32 / $3.96deepseek/deepseek-v4-pro-0813Source · |
| Claude Opus 5.5Anthropic | Not offered yet | $4.00 / $20.00anthropic/claude-opus-5.5Source · | $4.00 / $20.00anthropic/claude-opus-5-5Source · |
| Gemini 3.1 ProGoogle | Not offered yet | $2.00 / $12.00google/gemini-3.1-pro-previewSource · | $2.00 / $12.00google/gemini-3.1-pro-previewSource · |
What a month costs, fees included
Same model, same tokens, no cache hits. Fees on usage and on a card top-up for exactly what the tokens cost are added.
No model has a current price on every side today, so there is no like-for-like example to show. See the price table above.
Key differences
Resellers and gateways
OpenRouter, Requesty and FullsendAI sell model access: you pay them and they pay the providers. Portkey doesn't: it routes and logs calls made with your own provider keys and charges for the logs.
Privacy defaults
They differ more than the marketing suggests. OpenRouter: prompts not logged by default; opt in for a 1% discount. Requesty: self-serve: prompts and outputs logged for up to 30 days; can be turned off per key. Portkey logs requests by design; that's the product. FullsendAI stores API prompts only when you turn on body storage.
Ownership
Portkey now belongs to Palo Alto Networks and is sold as part of its Prisma AIRS platform. Unify's router API no longer answers; its site now sells continual learning for agents.
Migrate in 2 minutes
Change the base URL, the key and the model name. Everything else stays: streaming, tool calls and images work as before.
from openai import OpenAI
client = OpenAI(
base_url="https://fullsendai.com/v1",
api_key="fs_live_...",
)
reply = client.chat.completions.create(
model="deepseek-v4.1-flash",
messages=[{"role": "user", "content": "Hello"}],
)Full setup for every tool is in the quickstart and agent guides.
Questions
Which LLM gateway has the lowest fees?
On the fees each publishes, checked Oct 15, 2026: FullsendAI charges none on usage or top-ups; Requesty adds 5% to model cost; OpenRouter charges 5.5% by card ($0.80 minimum) and 5% in crypto on credit purchases; Portkey charges per log, not per token. Per-model prices differ too; see the table.
Is Portkey still independent?
No. Palo Alto Networks completed its acquisition of Portkey in May 2026, and Portkey's pricing page now presents it as Prisma AIRS AI Gateway. The source is linked in the table above.
What happened to Unify?
Its router API returned HTTP 404 when we checked, and its website now sells continual learning for production agents.
Where does FullsendAI run models?
In the United States, every model. Requests are never routed abroad, not even when a host is busy.
Does FullsendAI keep my prompts?
Not from the API unless you ask: by default only what billing needs (tokens, cost, timing, status). If you turn on body storage for a project, its requests and replies are kept for 30 days so you can read them in Logs, then deleted.
Related comparisons
One key. Every model. No top-up fee.
Works with the OpenAI and Anthropic SDKs, Claude Code, Codex and other coding agents. Every model runs in the USA.