DeepSeek V4 Flash is free to use until Aug 12. Get an Endpoint →
    Built by the team that solved sub-second cold starts for large language models.

    Inference infrastructure for production AI.

    Run AI models in production without paying for idle GPUs.

    True serverless inferenceSub-second cold startsOpenAI-compatible APIs
    Start BuildingExplore Models

    A request enters the OpenAI-compatible API, the scheduler restores initialized model state, GPU capacity attaches while tokens stream, and resources return to the pool when serving completes.

    INFERX RUNTIMEOPENAI-COMPATIBLE API
    GPU SNAPSHOT READY
    RESOURCE LAYERCAPACITY IDLE
    Snapshot PoolInitialized model state
    GPU PoolGPU slicing enabled0 SLICES
    Secure RuntimeIsolated container
    Scale to ZeroCapacity released

    Compatible with your favorite AI tools

    OpenAILangChainOpenCodeHermesOpenClawContinueLlamaIndexVercel AI SDKDify
    Compatibility references do not imply partnership or endorsement.

    THREE WAYS TO BUILD

    Choose how you run inference.

    Start with a hosted API, deploy your own model serverlessly, or run the complete InferX Platform in your infrastructure.
    READY-MODEL APIs

    InferX Endpoints

    Access production-ready open models through OpenAI-compatible APIs. InferX manages the infrastructure, scaling, and runtime.

    • Start with an API key
    • Pay per token
    • No infrastructure to manage
    Explore Endpoints
    CUSTOM MODELS

    True Serverless Inference

    Deploy your own model on InferX and pay only for the GPU compute used while serving requests.

    • Scale to zero when idle
    • No idle GPU cost
    • Sub-second cold starts
    • OpenAI-compatible APIs
    Deploy a Custom Model
    YOUR INFRASTRUCTURE

    InferX Platform

    Deploy the complete InferX inference platform on Kubernetes, bare metal, private cloud, GPU cloud, or air-gapped infrastructure.

    • Infrastructure remains under your control
    • Multi-tenant inference platform
    • Enterprise deployment support
    • VM-level workload isolation
    Explore Platform DeploymentView Architecture

    SECURITY POSTURE

    Enterprise-ready infrastructure.

    View Security
    Zero data retention

    Zero data retention

    Prompts, outputs, and API payloads are not stored or used for training.

    Encrypted in transit

    Encrypted in transit

    Requests and responses are protected with modern transport encryption.

    SOC 2 Type II

    SOC 2 Type II

    In progress, targeted for Q3 2026.

    HIPAA / GDPR

    HIPAA / GDPR

    Readiness targeted for Q3 2026.

    Production-ready endpoints.

    Start with hosted endpoints, or deploy more models from the InferX catalog.

    Ready now

    DeepSeek V4 Flash

    FREE TO USE UNTIL AUG 12

    DeepSeek

    Input
    $0.07 / 1M50% off
    Cached input
    $0.01 / 1M
    Output
    $0.11 / 1M61% off

    Cached input supported.

    Use endpoint
    Coming shortlyNEW

    DeepSeek-V4-Flash-0731

    DeepSeek

    Cached input
    $0.0028 → $0.0014 / 1M50% lower
    Output
    $0.28 → $0.14 / 1M50% lower

    Cached input supported.

    • Upgraded agent and coding capabilities
    • Native Responses API support
    • Codex compatible
    Coming Shortly
    Ready now

    Mimo v2.5

    Xiaomi

    Input
    $0.08 / 1M43% off
    Cached input
    $0.008 / 1M
    Output
    $0.18 / 1M36% off

    Cached input supported.

    Use endpoint
    Ready now

    GLM 5.2

    Zhipu AI

    Input
    $0.18 / 1M87% off
    Cached input
    $0.018 / 1M
    Output
    $1.90 / 1M57% off

    Cached input supported.

    Use endpoint
    Coming soon

    Kimi K3

    Moonshot AI

    Coming soonPricing will be published before availability.

    Introductory pricing compared with common market API rates.

    Explore more models deployable from the InferX catalog

    Our sub-sec cold starts, proven in production.

    MODEL
    Qwen3.6-35B-A3B
    PRECISION
    FP16
    MODEL SIZE
    ~70 GB
    HARDWARE
    NVIDIA H100
    TRADITIONAL STARTUPLoad model weights≈35s
    1. 01Pull container image9.5s
    2. 02Init runtime3s
    3. 03Load model weights17.5s
    4. 04Warm CUDA graphs3.5s
    5. 05Health check1.5s
    INFERX STARTUPGPU Snapshot Restore<1s
    1. 01Restore GPU snapshot<1s
    2. 02Rebind device memoryfast path
    3. 03Ready to serve0s idle

    SHORTER STARTUP PATH

    1. Request
    2. Snapshot restore
    3. Scale to zero

    Inside the InferX Runtime

    A model-ready runtime, restored on demand.

    Explore the complete architecture

    A request enters the OpenAI-compatible API, the scheduler restores initialized model state, GPU capacity attaches while tokens stream, and resources return to the pool when serving completes.

    INFERX RUNTIMEOPENAI-COMPATIBLE API
    GPU SNAPSHOT READY
    RESOURCE LAYERCAPACITY IDLE
    Snapshot PoolInitialized model state
    GPU PoolGPU slicing enabled0 SLICES
    Secure RuntimeIsolated container
    Scale to ZeroCapacity released

    OpenAI compatible

    Change one line. Keep your OpenAI code.

    Keep the OpenAI Python SDK and request shape. Point your existing client at InferX to serve open models in production.

    Verified SDK compatibility
    Get API KeyRead Docs
    inference.py
    PYTHON 3.12
    import osfrom openai import OpenAIclient = OpenAI(    api_key=os.environ["OPENAI_API_KEY"],    base_url="https://model.inferx.net/v1",)stream = client.chat.completions.create(    model="Qwen/Qwen3.6-35B-A3B",    messages=[{"role": "user", "content": "Hello, InferX."}],    stream=True,)for chunk in stream:    print(chunk.choices[0].delta.content or "", end="")
    InferX multi-tenant secure container runtime diagram showing guest user, guest kernel, host user, host kernel, and NCCL security isolation layers.

    Put production inference on a runtime built for it.

    Start BuildingContact Sales