Best API gateways for circuit breaking in microservices

Combined ranking

1. 1
   Kong GatewayGrade ↗Visit ↗incumbent—6 pts

   Claude #3Gemini #3

   The strongest all-round API gateway proper, with active/passive health checks and upstream ejection plus a deep plugin ecosystem, mature multi-platform ops, and a clear management story for teams that want a gateway rather than a mesh.

   + model takes & fixes− hide details

   Claude The strongest all-round API gateway proper, with active/passive health checks and upstream ejection plus a deep plugin ecosystem, mature multi-platform ops, and a clear management story for teams that want a gateway rather than a mesh.

   Gemini (Near-tie with APISIX) Battle-tested upstream ring-balancer circuit breaking that combines active and passive health checks with automatic target ejection, reinforced by massive ecosystem adoption, versatile deployment topologies (hybrid, DB-less, K8s Ingress), and extensive enterprise operational tooling.

   Where it falls short

   per Claude Native circuit breaking is coarser than Envoy's (health-check ejection rather than fine per-request thresholds); the richest resilience behavior leans on Enterprise/custom plugins.

   per Gemini Not for teams wanting advanced statistical sliding-window or percentile-latency breaker criteria without enterprise licensing, as fine-grained resilience policies often require commercial plugins or custom Lua scripting.

2. 2
   Envoy GatewayGrade ↗Visit ↗incumbent—5 pts

   Claude —Gemini #1

   Sets the industry standard for L7 resilience via Envoy's battle-tested C++ data plane, providing multi-dimensional circuit breaking controls (native Outlier Detection for consecutive 5xx/gateway errors with automatic host ejection, alongside strict connection, pending request, and retry limits) with negligible latency overhead under the Kubernetes Gateway API standard.

   + model takes & fixes− hide details

   Gemini Sets the industry standard for L7 resilience via Envoy's battle-tested C++ data plane, providing multi-dimensional circuit breaking controls (native Outlier Detection for consecutive 5xx/gateway errors with automatic host ejection, alongside strict connection, pending request, and retry limits) with negligible latency overhead under the Kubernetes Gateway API standard.

   Where it falls short

   per Gemini Not for teams needing simple, low-ceremony edge routing; its configuration mental model and operational footprint are excessively complex for non-Kubernetes or modest architectures.

3. 3
   Envoy ProxyGrade ↗Visit ↗—5 pts

   Claude #1Gemini —

   Best-in-class circuit-breaking primitives in the category — granular per-cluster thresholds (max connections, pending requests, requests, retries) plus outlier detection that ejects failing endpoints automatically; battle-tested at scale and the engine underneath most other entries here, so it sets the ceiling for what these features can do.

   + model takes & fixes− hide details

   Claude Best-in-class circuit-breaking primitives in the category — granular per-cluster thresholds (max connections, pending requests, requests, retries) plus outlier detection that ejects failing endpoints automatically; battle-tested at scale and the engine underneath most other entries here, so it sets the ceiling for what these features can do.

   Where it falls short

   per Claude Raw Envoy is a proxy, not a turnkey managed gateway — hand-written xDS/config is verbose and there's no built-in control plane or UI, so it's not for teams wanting a product out of the box.

4. 4
   TraefikGrade ↗Visit ↗incumbent—4 pts

   Claude #4Gemini #4

   Cleanest developer experience for the common case — a declarative CircuitBreaker middleware with readable expressions on latency, error ratio, and network errors, tight Kubernetes/CRD and label-based config, and near-zero setup.

   + model takes & fixes− hide details

   Claude Cleanest developer experience for the common case — a declarative CircuitBreaker middleware with readable expressions on latency, error ratio, and network errors, tight Kubernetes/CRD and label-based config, and near-zero setup.

   Gemini Features the most accessible and expressive native circuit breaker middleware in the category, allowing operators to trigger breakers using clean mathematical expressions (e.g., LatencyAtQuantileMS, NetworkErrorRatio, or ResponseCodeRatio) with automated service discovery across container runtimes.

   Where it falls short

   per Claude Less granular than Envoy (no per-endpoint outlier ejection or connection-pool shedding), so it's not for teams needing fine-tuned, high-scale failure isolation.

   per Gemini Not for hyper-scale distributed traffic requiring granular per-connection pooling limits, distributed breaker coordination across gateway nodes, or advanced outlier ejection in multi-cluster topologies.

5. 5
   Apache APISIXGrade ↗Visit ↗—4 pts

   Claude —Gemini #2

   (Near-tie with Kong) Offers a highly performant, dynamic built-in api-breaker plugin managing healthy, unhealthy, and half-open states with zero-reload updates via etcd; delivers exceptional throughput and low memory consumption while enabling granular status-code thresholds and upstream health checks.

   + model takes & fixes− hide details

   Gemini (Near-tie with Kong) Offers a highly performant, dynamic built-in api-breaker plugin managing healthy, unhealthy, and half-open states with zero-reload updates via etcd; delivers exceptional throughput and low memory consumption while enabling granular status-code thresholds and upstream health checks.

   Where it falls short

   per Gemini Not for teams looking for plug-and-play enterprise governance out of the box, as deep troubleshooting and complex custom breaker extensions require specialized Lua/WASM and OpenResty expertise.

6. 6
   IstioGrade ↗Visit ↗incumbent—4 pts

   Claude #2Gemini —

   The most practical way to get Envoy's full circuit breaking across a fleet, applied declaratively per-service via DestinationRule with connection-pool limits and outlier detection, with no app code changes; ambient mode has meaningfully cut the resource/ops cost that historically counted against it.

   + model takes & fixes− hide details

   Claude The most practical way to get Envoy's full circuit breaking across a fleet, applied declaratively per-service via DestinationRule with connection-pool limits and outlier detection, with no app code changes; ambient mode has meaningfully cut the resource/ops cost that historically counted against it.

   Where it falls short

   per Claude Still a full service mesh — CRD surface and operational overhead are overkill for a handful of services or non-Kubernetes shops.

7. 7
   SSolo.io Gloo GatewayGrade ↗—1 pts

   Claude #5Gemini —

   Exposes Envoy's complete circuit-breaking and outlier-detection feature set through a supported gateway/control plane with policy management and multicluster reach — Envoy power without hand-rolling xDS. Near-tie with Kong on overall gateway merit; ranked below it on breadth of adoption and ecosystem.

   + model takes & fixes− hide details

   Claude Exposes Envoy's complete circuit-breaking and outlier-detection feature set through a supported gateway/control plane with policy management and multicluster reach — Envoy power without hand-rolling xDS. Near-tie with Kong on overall gateway merit; ranked below it on breadth of adoption and ecosystem.

   Where it falls short

   per Claude Commercial with smaller community and cost/lock-in considerations; heavier than needed unless you specifically want managed Envoy at scale.

8. 8
   SSpring Cloud GatewayGrade ↗—1 pts

   Claude —Gemini #5

   Unmatched depth of resilience configuration within Java/Spring ecosystems via native integration with Resilience4j, providing sophisticated sliding-window failure rates, slow-call duration thresholds, and seamless code-level fallback handling.

   + model takes & fixes− hide details

   Gemini Unmatched depth of resilience configuration within Java/Spring ecosystems via native integration with Resilience4j, providing sophisticated sliding-window failure rates, slow-call duration thresholds, and seamless code-level fallback handling.

   Where it falls short

   per Gemini Not for polyglot infrastructure; its high JVM memory footprint and deep coupling to the Spring Framework make it unviable as a universal, language-agnostic network gateway.

Just missed the top 5

By model

Common questions

Tracked by ModelsAgree · rank 1 = 5 pts … rank 5 = 1 pt · re-polled on demand