LiteLLM vs SGLang
LiteLLM
litellm.aiSGLang
sglang.ioSide by side
| Figure | LiteLLM | SGLang |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | Yes | — |
| Open source | — | — |
| Web | Yes | — |
| macOS | — | Yes |
| Linux | — | Yes |
| API | Yes | Yes |
| Self-hosted | Yes | Yes |
| Plans published | 2 | 1 |
| Spec score · LLM Gateway Software | 7.8 (#15) | 8.1 (#12) |
| Spec score · LLM gateways | 7.8 (#19) | 8.4 (#11) |
| Caching | Response and semantic caching can use Redis, S3, or GCS. | The project describes hierarchical KV caching across GPU memory, host memory, and external storage through ecosystem projects including HiCache, Mooncake, and LMCache. |
| Integrations | The integrations documentation lists observability tools including Langfuse, Datadog, Grafana Cloud, OpenTelemetry, LangSmith, Arize/Phoenix, and Helicone. | The README lists deployment and orchestration integrations including Ray Serve, NVIDIA Dynamo, and llm-d. |
| Security | The product page describes signed hardened non-root images, vulnerability scanning, SSO/JWT/RBAC controls, and secrets-manager integrations. | The repository identifies its license as Apache-2.0. |
| Support | Enterprise includes dedicated support and onboarding, with 24/7 support and SLAs listed on the pricing page. | The project points users to GitHub issues, Slack, Discord, and community discussions for questions and help. |
Plans
In their own words
LiteLLM
LiteLLM is an open-source AI gateway and LLM proxy for platform teams. It provides one OpenAI-compatible API to 140+ providers and 1,800+ models, with load balancing, lowest-cost routing and automatic selection of cheaper models for simple prompts. Teams can track usage and spend across keys, users, teams, organizations, tools, agents and MCP, then apply hard budgets and rate limits. Response and semantic caching can use Redis, S3 or GCS. LiteLLM can be self-hosted, including in air-gapped environments, and offers deployment through an official Helm chart or Terraform module. The maker says self-hosting has no telemetry and keeps data within the user's infrastructure. Listed security features include hardened non-root images, vulnerability scanning, SSO/JWT/RBAC controls and secrets-manager integrations. The free Open Source plan is self-hosted and includes 100+ providers, virtual keys, budgets, rate limits, fallbacks, logging and Prometheus metrics. A 30-day Enterprise trial key is available without a credit card.
SGLang
SGLang is a free, open-source inference framework for serving large language, vision-language and diffusion models. It targets agentic workloads, reinforcement-learning rollouts and large-scale serving, from one GPU to distributed clusters. Runtime features include RadixAttention, prefix caching and multi-GPU parallelism. Listed optimizations include disaggregated prefill and decode, speculative decoding, a zero-overhead scheduler and optimized GPU kernels. Supported models include DeepSeek, Qwen, GPT-OSS, Llama, Mistral and GLM. SGLang works with Hugging Face and OpenAI APIs and exposes OpenAI-compatible endpoints. It installs with pip or Docker and supports self-hosted deployment. Hardware support spans NVIDIA and AMD GPUs, Google TPUs, Intel GPUs and CPUs, Apple Silicon, Huawei Ascend NPUs and Moore Threads GPUs. SGLang Diffusion adds image and video generation. The Apache-2.0 license permits open-source use.