LLM gateway comparison
Last updated: October 23, 2026.
LLM gateways sit between your applications or agents and the model providers, so you can route calls, cap spend, restrict models and log every request in one place. They come from four directions. Open-source proxies (LiteLLM, Helicone) run in your own infrastructure. Managed routers (OpenRouter, Requesty, Vercel AI Gateway, Cloudflare AI Gateway) give one account across many providers. API gateways add AI plugins to existing API management (Kong, Azure API Management), and AI platforms pair the gateway with deployment and guardrails (Portkey, TrueFoundry). Mintive's LLM gateway (early access) governs agents' model traffic with the same users, groups and audit trail as its MCP gateway, so tool calls and model calls follow one set of policies.
How we compared these LLM gateways
The facts come from each vendor's public documentation, pricing pages and repositories, checked on October 23, 2026. A cell says "not documented" when we couldn't find the fact on the vendor's own pages, which doesn't mean the capability is missing. There is no ranking column: each row ends with when it fits best.
LLM gateway pricing, deployment and supported providers
| Gateway | Type | Deployment and license | Pricing model | Providers and models | Best fit when |
|---|---|---|---|---|---|
| Mintive | LLM gateway beside the MCP gateway and Agent Monitor (early access) | Managed cloud service in the US or EU; Private Networks reach internal MCP services (Enterprise); on-premise or self-hosted by arrangement. License: paid plans | Early access; contact Mintive | OpenAI, Anthropic, Fireworks and Amazon Bedrock (providers) | You want agents' model spend and access governed with the same identities as their MCP tools |
| LiteLLM | Open-source LLM gateway (proxy) | Self-hosted. License: MIT except the enterprise/ directory | Open source $0; Enterprise annual | 100+ LLMs | You want an open-source proxy you run and extend yourself |
| Portkey | LLM gateway and control plane | SaaS, self-hosted, private cloud or VPC. License: Gateway repository MIT; platform paid | Free; Production $49/month; Enterprise custom | 1,600+ LLMs | You want a managed gateway with built-in guardrails and workspaces |
| OpenRouter | Managed model router | Managed; EU endpoint. License: commercial | 5.5% fee on credits (8% on Business); bring-your-own-key options | 500+ models from 80+ providers | You want one account across many providers and models |
| Kong AI Gateway | API gateway with AI plugins | Konnect hybrid or serverless; self-hosted. License: Kong Gateway Apache 2.0; AI plugins on Enterprise | Plus from $25/month plus usage | Major providers through one API | You already run Kong for APIs |
| Cloudflare AI Gateway | Managed AI gateway | Managed, on all Cloudflare plans. License: commercial | Core features free; 5% fee on unified billing credits | Workers AI, OpenAI, Anthropic, Gemini and more | You run on Cloudflare and want a free managed gateway |
| TrueFoundry | AI gateway on an AI deployment platform | SaaS, hybrid, self-hosted VPC or on-premise. License: commercial | Free (3 users); Pro $25 per user/month; Enterprise custom | 1,000+ LLMs | You want a self-hostable gateway beside model and agent deployment |
| Helicone | Open-source observability with an AI gateway | Cloud or self-hosted. License: Apache 2.0 | Free; Pro $79/month; Team $799/month | 100+ providers | You mainly want open-source request logging and cost tracking |
| Vercel AI Gateway | Managed AI gateway | Managed, callable from anywhere. License: commercial | No token markup; paid add-ons | 400+ models from 70+ providers | You want no-markup model access with one-command coding-agent setup |
| Azure API Management | AI gateway features in API Management | Managed Azure; self-hosted gateway. License: commercial | API Management tier pricing | OpenAI-compatible, Anthropic, Vertex; Foundry and Bedrock models | Your models run in Azure and you already use API Management |
| Requesty | Managed model router | Managed; EU data residency. License: commercial | Free; 5% markup pay as you go | 600+ models | You want managed routing with organization model allowlists |
Portkey's site states that it is now part of Palo Alto Networks as "Prisma AIRS AI Gateway", and Helicone's site states that it is no longer part of Mintlify.
LLM gateway budgets, access control and Claude Code support
| Gateway | Subscription seats | Budgets and rate limits | Allowed models per group | Identity (SSO, SCIM, groups) | Coding-agent setup documented |
|---|---|---|---|---|---|
| Mintive | Anthropic: routes through each caller's linked Claude subscription | A monthly budget per policy, with optional per-user caps and per-user overrides; calls get a 429 once a budget is spent | Yes: models per policy, policy granted to groups | SSO and SCIM; Can Use grants by group or user (access) | Claude Code, Codex and Claude Desktop through the mintive CLI; Coworker Agents |
| LiteLLM | Claude Pro, Max, Team and Enterprise | Per key, user, team and organization | Per key; model access groups | SSO and SCIM on Enterprise | Claude Code, Codex, Cursor |
| Portkey | Claude Max and Anthropic OAuth | Per workspace, team and key | Yes, per workspace | SSO and SCIM | Claude Code, Codex, Cursor and more |
| OpenRouter | Not documented | Per member, key and workspace | Yes, per member, key and workspace | SAML SSO and SCIM on Enterprise | Claude Code, Codex, Cursor |
| Kong AI Gateway | No: its Claude Code guide sends seats directly to Anthropic | Token and cost limits per consumer group | Yes, per consumer group | SSO, SCIM and RBAC on Enterprise | Claude Code; Codex CLI mentioned |
| Cloudflare AI Gateway | Not documented | Spend limits by model, provider or user | Not documented | Cloudflare Access with your IdP | Claude Code, Codex, Cursor |
| TrueFoundry | Claude plans; ChatGPT plans for Codex | Per user, team, virtual account and model | Yes, per team or user | OIDC or SAML SSO and SCIM | Claude Code, Codex, Cursor |
| Helicone | Not documented | Rate limits per user or property, cost-based | Not documented | SAML SSO on Enterprise | Claude Code, Codex |
| Vercel AI Gateway | Claude subscription with Claude Code | Per team, project, key and member | Team-wide allowlist | SAML SSO; directory sync on Enterprise | Claude Code, Codex, Cursor and 25+ more |
| Azure API Management | Not documented | Token limits per subscription or custom key | Not documented | Entra ID or another IdP over OAuth | Not documented |
| Requesty | Not documented | Per key, user and group | Yes: organization list plus group access lists | SAML or OIDC SSO; SCIM on request | Claude Code, Codex |
LLM gateway logging, guardrails and compliance
| Gateway | Logging and audit export | Guardrails on LLM traffic | Compliance stated by the vendor |
|---|---|---|---|
| Mintive | Every call logged in Agent Monitor; budget changes in the audit log; SIEM export over OTLP or Splunk HEC, or as batched files to S3 | No content screening in the LLM gateway itself; Agent Monitor rules screen coding-agent prompts for secrets and prompt injection through agent hooks; Mint Guard (enabled on request) screens MCP tool calls | SOC 2 Type II and HIPAA, with a BAA available; ZDR covers the LLM gateway |
| LiteLLM | OpenTelemetry, Datadog, S3; audit logs on Enterprise | Presidio PII masking; Lakera, Bedrock Guardrails and others | SOC 2 Type II |
| Portkey | OpenTelemetry export; audit log | PII detection; partner prompt-injection checks | SOC 2, ISO 27001, GDPR, HIPAA |
| OpenRouter | Broadcast to OpenTelemetry, Datadog, S3 and more | PII redaction; prompt-injection detection | SOC 2 Type II |
| Kong AI Gateway | Audit logs, metrics, OpenTelemetry | PII sanitizer; prompt guard | Not documented on the pages checked |
| Cloudflare AI Gateway | Logs, Logpush, OpenTelemetry traces | Safety guardrails; DLP | Not documented on the AI Gateway pages |
| TrueFoundry | OpenTelemetry; audit logs | PII redaction; prompt-injection detection | SOC 2, HIPAA, GDPR |
| Helicone | Request logs | Prompt Guard (OpenAI models) | SOC 2 Type II, HIPAA |
| Vercel AI Gateway | Request logs; OpenTelemetry trace drains | None at the gateway level, per Vercel | SOC 2 Type 2, ISO 27001, HIPAA BAA |
| Azure API Management | Azure Monitor and Application Insights | Azure AI Content Safety with prompt shields | Not documented on the AI gateway pages |
| Requesty | Usage analytics | PII and secrets scanning on Enterprise | Not documented |
How to choose an LLM gateway
LLM gateway for AI agents and coding agents
Coding agents and autonomous agents make model calls and tool calls in the same session, so governing only one half leaves gaps. Look for budgets and allowed models tied to your directory groups, setup for the harnesses your people run (Claude Code, Codex, Cursor), and logs that land in the same place as tool-call logs. Mintive's LLM gateway uses the same users and groups as its MCP gateway and logs every call to Agent Monitor, and the mintive CLI configures local harnesses (user setup).
Open source and self-hosted LLM gateways
LiteLLM (MIT, with an enterprise directory) and Helicone (Apache 2.0) run in your own infrastructure, and Portkey's gateway repository and Kong Gateway are open source too. You operate upgrades, scaling and identity integration yourself, and SSO, SCIM and audit logs are often on paid tiers.
LLM routers for many models and providers
Managed routers such as OpenRouter, Requesty, Vercel AI Gateway and Cloudflare AI Gateway give one API and one bill across hundreds of models. Compare their fees (a percentage on credits, or no markup), their model allowlists, and whether budgets can be set per team or member.
AI gateway features in an existing API gateway
Kong AI Gateway and Azure API Management add token limits, content safety and model routing to the API management you already operate. Check how each handles subscription seats and coding agents, since Kong's Claude Code guide sends subscription traffic directly to Anthropic.
LLM gateway FAQ
What is the difference between an LLM gateway and an AI gateway?
The terms overlap. Kong, Cloudflare, Vercel and Azure call their products AI gateways, and they route LLM traffic alongside other AI traffic, with Kong and Azure also covering MCP. An LLM gateway focuses on model calls: provider routing, budgets, allowed models and request logs. See the MCP gateway comparison for tool traffic.
Is there an open source LLM gateway?
Yes. LiteLLM is MIT licensed except its enterprise directory, Helicone is Apache 2.0, Portkey's gateway repository is MIT, and Kong Gateway is Apache 2.0 with its AI plugins on the Enterprise tier. You host and operate these yourself, and SSO, SCIM and audit logs are often paid features.
Which LLM gateways work with Claude Code?
Mintive, LiteLLM, Portkey, OpenRouter, Kong, Cloudflare, TrueFoundry, Helicone, Vercel and Requesty all document Claude Code setup. Mintive's CLI configures local harnesses such as Claude Code and Codex, and Vercel's CLI configures Claude Code, Codex, Cursor and other agents in one command.
Can an LLM gateway use Claude or ChatGPT subscription seats instead of API keys?
Some can. Mintive (Anthropic), LiteLLM, Portkey, TrueFoundry and Vercel document routing Claude subscription traffic, and TrueFoundry also documents ChatGPT plans for Codex. Kong's Claude Code guide sends subscription seats directly to Anthropic instead. See provider and model selection.
How do LLM gateways limit spend by team?
Most attach budgets to a unit you manage: a policy, with optional per-user caps (Mintive), a key, team or organization (LiteLLM), a workspace (Portkey, OpenRouter), a consumer group (Kong), or a team, project or key (Vercel). Azure API Management limits tokens per subscription or custom key rather than dollars. See budgets.
Do LLM gateways screen prompts for PII and prompt injection?
Many do, with different coverage. LiteLLM, Portkey, OpenRouter, TrueFoundry, Kong and Azure document PII or prompt-attack checks on LLM traffic, and Vercel states its gateway ships no guardrails. Mintive screens prompts from coding agents with Agent Monitor rules and screens MCP tool calls with Mint Guard.
Related
- LLM gateway: route agent inference through Mintive
- MCP gateway comparison: 13 MCP gateways side by side
- AI agent monitoring tools comparison
- Best LLM gateways on the blog
Sources
Every competitor fact on this page comes from that vendor's own public documentation, pricing pages or repository, checked on October 23, 2026.