Inference as a Service

Production
Inference Stack

End-to-end inference with automatic scaling, streaming tokens, and real-time monitoring.

View ModelsOpen DeployPad
Infrastructure

Deployment Modes

Choose the optimal runtime for your workload. From elastic API endpoints to bare-metal isolated instances.

Serverless and dedicated deployment comparison
Compare by01 / Serverless

Serverless Inferencing

AUTO_SCALING		02 / Dedicated

Dedicated Deployment

ISOLATED_RUNTIME
Workload fit

Serverless endpoints are optimized for rapid deployment and elastic workloads. Fully abstracted infrastructure with automatic scaling.

Used when workload predictability, isolation, or sustained throughput becomes critical. Single-tenant GPU allocation.

ExecutionMulti-tenant optimized runtimeSingle-tenant isolated runtime
InfrastructureFully abstractedDedicated GPU allocation
ControlParameter-level tuningInfra + runtime control
ScalingAutomatic, workload-drivenCluster-level scaling
Use CaseDynamic workloads, API productsStable high-throughput systems
Get started
Start with ServerlessServerless Inferencing Docs
Configure Dedicated DeploymentDedicated Deployment Docs
Available Models

Run the models you need.

01 / Model Type: Text

GLM 5.3 Flash

GLM 5.3 Flash is well suited to coding agents, repository-scale debugging, tool-using workflows and visual tasks such as understanding screenshots, charts and documents. Use it when an agent needs to work across a large codebase or long research context, call tools, inspect visual inputs and continue through multi-step tasks without switching models. Z.ai positions the model particularly around coding, multimodal understanding and agentic work.

zai-org/glm-5.3-flash
$0.110per 1M tokens
02 / Model Type: Text

DeepSeek V4.1 Flash

DeepSeek V4.1 Flash is built for fast coding, software-engineering agents and high-volume agentic workloads that still require strong reasoning. It is a good fit for terminal automation, code debugging, tool calling, research agents, document or screenshot analysis, and long-running workflows that repeatedly reuse context. For new DeepSeek Flash integrations, V4.1 is the newer multimodal generation and should normally replace the original V4 Flash.

deepseek-ai/DeepSeek-V4.1-Flash
$0.300per 1M tokens
03 / Model Type: Text

openai/gpt-oss-120b

GPT-OSS-120B is OpenAI's higher-capability open-weight model for complex reasoning, coding agents, function calling and structured outputs. Use it for workloads that require multi-step problem solving, planning, tool use or more difficult coding tasks, especially when you want an open model that can be customized or deployed outside OpenAI's hosted API.

openai/gpt-oss-120b
$0.039per 1M tokens

Explore Full Library
How it works

Better kernels.
Better inference.

Our hardware-specific LLMs help agents write better kernels. We test the changes and return verified improvements to the inference service.

  1. 01 / Runtime

    Measure performance.

    Runtime performance measurements show where execution can improve. Our optimization agents do not inspect customer prompts or completions.

  2. 02 / Model Engine

    Write and test kernels.

    Meridian, our NVIDIA-specific LLM, powers AI agents that write and refine CUDA kernels. The fully automated workflow tests correctness and execution performance, then deploys verified improvements to our inference service.

  3. 03 / Inference service

    Deploy and improve.

    Verified improvements return to serving. Kernel code and measured results guide the next cycle, not customer content.

Our optimization loop does not read or retain your prompts or completions, or use them for model training.

Explore Model Engine
Infrastructure

Built for continuous operation.

Regional GPU capacity, redundant infrastructure, and direct engineering ownership support production inference under sustained load.

NVIDIA GPUs
500+
Inference capacity across multiple regions.
Observed uptime
99.95%
Across multi-location deployment with failover.
Datacenter infrastructure
Tier III
Redundant power, network, and hardware.
Failure handling

Resilience in the stack.
Engineers in the loop.

Infrastructure and MLOps work together, from failure detection to recovery. Incidents go directly to the engineers responsible for the runtime.

  1. Reduce single points of failure

    Redundant power, network, and hardware underpin the serving infrastructure.

  2. Detect and recover

    Failure detection and automated failover support continuous operation across locations.

  3. Respond directly

    Engineers receive incident alerts directly. Infrastructure and MLOps coordinate recovery and runtime tuning.

Compliance

Enterprise AI Inference,
Built for GDPR.

Deploy production-ready AI models without the compliance headache. Geodd provides high-performance, GDPR-ready inference designed to eliminate unnecessary data exposure for products serving the EU, EEA, and UK.

  • Zero-Data Retention

    We never store your API prompts, completions, request bodies, or customer datasets.

  • No Training or Human Review

    Your data is strictly yours. It is never used for model training or human-in-the-loop screening.

  • EU Data Sovereignty

    Route and process your inference workloads entirely within EU-based data center infrastructure.

  • Enterprise Security

    Hardened with encryption at rest and in transit (TLS), role-based access controls, and a comprehensive DPA framework.

  • The Geodd Standard

    We act strictly as a Data Processor for your API data, keeping your internal workflows, product logic, and customer messages safe by default.

Ready to Scale?

Join the next generation of AI

Build on Geodd's hyper-optimized inference stack. Get instant API access to the world's most capable open-source models or talk to our team for custom deployments.

Get API AccessTalk to Sales