Model Gateway
One OpenAI-compatible interface to integrated models and their providers. Keep the request familiar. Choose how the work gets served.
POST /v1/chat/completions“Summarize this incident report.”
Keep the model. Change the preference.
Provider routing chooses among currently available providers compatible with your request. It does not replace the model you asked for.
Balances price, speed and quality across eligible providers.
"routing_preset": "Balanced"Selects the provider with the lowest applicable input and output token price—not a guaranteed lowest total application cost.
"routing_preset": "Cheapest"Prioritizes provider throughput. This is not a fixed latency promise or a guarantee of the shortest end-to-end request.
"routing_preset": "Fastest"Chooses by AIVAX's provider quality ranking, without optimizing for price or speed. It does not select a different model.
"routing_preset": "Quality"Explanations of routing behavior, not a live route simulation. No provider prices, speed measurements or rankings are fabricated here.
A different question: which model?
Configure a model for each complexity level in an AI Gateway. The complexity router classifies the latest user request as low, medium or high, then selects the model configured for that level.
Provider routing chooses where a model runs. Complexity routing can change which model runs. Keep those decisions distinct.
Configure model routingConceptual mapping, not a performance tier or benchmark. When available, X-Model-Routed-Complexity reports the selected complexity level on the HTTP response.
A familiar call, with a deliberate choice
Use an integrated model tag for a direct call. Use an AI Gateway ID or private-key slug when the model configuration, instructions, knowledge and tools should be shared.
Save the preference in your AI Gateway as parameters.routingOption. A request-level routing_preset overrides it for that call only. It requires a private API key and is an AIVAX extension to the OpenAI-compatible request body.
Your backend submits the report once. AIVAX selects an eligible provider for the requested integrated model using the preference you supply.
Send to POST /v1/chat/completions with Bearer authentication. Replace the model placeholder with an available integrated model tag or a gateway using one.
{
"model": "YOUR_INTEGRATED_MODEL_OR_GATEWAY_ID",
"messages": [{
"role": "user",
"content": "Summarize this incident report: checkout requests timed out after the database pool reached capacity."
}],
"routing_preset": "Cheapest"
}This example shows request structure. No inference is executed on this page.
From the model catalog
186 available models across 34 catalog groups. Snapshot: Sep 21, 2026. Groups identify model families, not the number of serving providers.
Compare all models@z-ai/glm-5.3-flashx
GLM-5.3-FlashX is the high-speed variant of Z.ai's GLM-5.3-Flash, a native multimodal model delivering inference speeds of up to 200 tokens/s. Built on the same hybrid sparse and linear attention architecture (320B total parameters, 18B active), it is suited for efficient coding, visual understanding, and long-horizon agent tasks with a 1M-token context window.
@unbiased/pareto
Pareto is a multimodal composite model built for research, coding, and agentic workflows, while delivering frontier-level performance across a broad range of general-purpose tasks.
@sakana/fugu-max
Fugu Max is the cost-performance model in Sakana AI's Fugu family. Rather than a single monolithic model, Fugu is a learned multi-agent orchestration system: a language model trained to route tasks across a fixed pool of open-weights and specialized models, with a 1M-token context window.
Latest dated releases among available models, ordered by the catalog's release date—not the date AIVAX added them. Refreshed when this site is built; this is not a live availability or latency guarantee.