Model catalog

96 models.
17 providers.
One endpoint.

Integrate once, then route each call to whichever provider fits it. One OpenAI-compatible endpoint reaches every provider in the catalog.

OpenAI-compatible17 providersLive catalog, not a static list
The network

17 providers, one integration

Every provider below sits behind the same endpoint and the same API key. Switching between them is a change of model slug, nothing more.

OpenAI

GPT-class frontier models

Anthropic

Claude family built for long-context reasoning

Google Gemini

Multimodal Gemini models

DeepSeek

V3 chat + R1 reasoning

Qwen (Alibaba)

Qwen3 across sizes + coder variants

Mistral

Open-weight European models

Groq

LPU inference built for very low latency

Fireworks AI

Broad open-weight hosting

DeepInfra

Cost-efficient open-weight hosting

Cerebras

Wafer-scale inference throughput

Cohere

Command models + retrieval

GLM / Zhipu AI

GLM-4 / GLM-5 family

Kimi / Moonshot

Long-context Kimi models

MiniMax

Long-context generation

Perplexity

Search-grounded answer models

StepFun

Step series models

xAI (Grok)

Grok reasoning models

Looking for a specific model? Browse the live catalog →

Equivalence groups

Same model, many providers

Some models run on more than one provider. Hober groups those by weights, not by vendor, so a routing decision can shift to whichever host is fastest or cheapest right now without changing which model answers the call.

DeepSeek V3chat
DeepSeekFireworks AIDeepInfra
DeepSeek R1reasoning
DeepSeekDeepInfra
Llama 3.1 8Bchat
Fireworks AIGroqDeepInfra
Llama 3.3 70Bchat
Fireworks AIGroq
Qwen3 235Bchat
Qwen (Alibaba)Fireworks AI
Qwen3 32Bchat
Qwen (Alibaba)Groq
GPT-OSS 120Bchat
Fireworks AICerebras
Mistral Smallchat
MistralDeepInfra

See how equivalence keys off routing decisions on the verifiable routing page →

Capability tiers

Route by what the work needs

Pick a tier yourself when you already know what the job needs, or leave the pick to Hober Auto Routing and let it decide per call.

Flagship

Top-tier reasoning and generation: best quality, higher cost. R1, Qwen3 235B, Kimi K2.

Workhorse

Strong general-purpose models that hit the quality/cost sweet spot. DeepSeek V3, Qwen3 Coder, GLM-4.

Economy

Lightweight and fast, built for the lowest cost on high-throughput workloads. Qwen3 32B, GLM-4 Flash.

Routing modes

Four ways to pick a model

Append a suffix to any model slug to choose what HAR (Hober Auto Routing) optimizes for on that call.

Auto
:auto
Optimizes for best overall.
Quality
:quality
Optimizes for highest quality.
Cost
:cost
Optimizes for lowest cost.
Fast
:fast
Optimizes for lowest latency.
For developers

Plug your agent in

One endpoint, 96 models, 17 providers, OpenAI-compatible.

@hober/sdk (coming to npm)
Ship on the catalog

Route your first request.

The catalog above is live right now. Get a key and call any model in it, or hand the pick to Auto Routing.

Pick a model or let Auto Routing choose for you.

Get an API key →