AI Inference

We use real workloads to build faster inference.

Run the models you need through one API.

Start building Explore model engine

Our AI agents monitor live inference performance, identify bottlenecks, and write better hardware kernels, the low-level code that runs your models.

AI agents use our own LLMs, fine-tuned for each accelerator. We test their changes against real workload conditions and deploy verified improvements into our inference service.

How It Works02

Production traffic guides every improvement.

  1. 01

    Observe real workloads

    Execution graphs, context lengths, and changing batch sizes show where inference spends time.

  2. 02

    Write better kernels

    Our AI agents use hardware-specific LLMs to generate and refine code for those bottlenecks.

  3. 03

    Test against the workload

    Check correctness and measure whether the changes improve inference under the conditions that exposed the problem.

  4. 04

    Deploy and measure

    Release verified improvements into serving, then monitor their effect on production performance.

  5. 05

    Feed the next cycle

    Successful kernels and measured results inform the next experiments and further development of our hardware-specific LLMs.

Our Kernel Development Models03

The models powering
this process.

Mosaic, Druze, and Strata power the agents in this process. Each is fine-tuned for the hardware it optimizes.

01 / CUDA

Mosaic · NVIDIA

Kernel development for NVIDIA GPUs and CUDA.

View benchmarks
Parallel / OrderedCUDA
02 / ROCm

Druze · AMD

Kernel development for AMD GPUs and ROCm.

Coming soon
Branching / AdaptiveROCm
03 / TT-Metalium

Strata · Tenstorrent

Kernel development for Tenstorrent accelerators and TT-Metalium.

Coming soon
Mesh / DistributedTT-Metalium
Pricing & Deployment

Serverless Inference

GLM 5.2

Model Type
Text
Quantization
fp8

We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include:

zai-org/glm-5.2
$0.900
per 1M tokens

DeepSeek V4 Flash

Model Type
Text
Quantization
fp8

DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.

deepseek-ai/DeepSeek-V4-Flash
$0.140
per 1M tokens

openai/gpt-oss-120b

Model Type
Text
Quantization
bf16

Welcome to the gpt-oss series, OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases.

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

Explore All Models

Multi-Regional

Deploy across 3 US regions, with 2 more continents coming soon.

GDPR Ready & SOC 2 Pending

Enterprise-grade security and data isolation for all workloads.

Unified API

One SDK for both serverless inference and dedicated compute.

Locations Topology

Built for Scale

Select a region

Select a location dot or region to inspect capacity

Select a region
Deployment footprint

Global infrastructure

Regional compute capacity positioned for production inference workloads.

3
Global
regions
2
Active
regions
Active
Expansion
Blog

Latest Updates

Geodd EU Serverless Inference Is Now Live
Uncategorised

Geodd EU Serverless Inference Is Now Live

Geodd EU serverless inference is now live, starting with GPU infrastructure hosted in Norway. EU customers can now run supported models through Geodd ...

Dedicated GPU vs Dedicated AI Inference
Uncategorised

Dedicated GPU vs Dedicated AI Inference

A dedicated GPU gives your team reserved GPU compute. Dedicated AI inference gives your team a dedicated or isolated inference environment that may in...

Total Cost of Self-Hosted Inference
Uncategorised

Total Cost of Self-Hosted Inference

The total cost of self-hosted inference is not only the hourly GPU price. It includes GPU capacity, supporting compute, storage, networking, orchestra...

Integration

Developer-First
Control

Fully compatible with the OpenAI SDK. Switch providers with a single line of code. No migration headaches, just immediate performance gains.

  • Direct OpenAI SDK compatibility
  • Real-time token usage and observability
  • Privacy first with Zero Data Retention (ZDR) and logging policy
deploy_inference.py
from openai import OpenAI

# Switch to Geodd by changing base_url
client = OpenAI(
  api_key="GEODD_API_KEY",
  base_url="https://api.geodd.io/inference/v1"
)

completion = client.chat.completions.create(
  model="openai/gpt-oss-120b",
  messages=[
    {"role": "user", "content": "What is machine learning?"}
  ]
)

print(completion.choices[0].message)

Ready to Scale?

Explore Geodd
Today.

Get instant access to our Model APIs and dedicated GPUs. Precision engineered for the most demanding production workloads.

Get Started Talk to Sales
deploy — mistral-large
Provisioning GPUdone
Preparing Serverdone
Configuring Accessdone
Downloading Modelrunning…
Starting Model Serverpending
Finalizing Servicepending
$awaiting pipeline