What's new in model router in Microsoft Foundry Models

This article lists model router releases and major documentation updates, newest first. Check it for new models, retirements, and features that change how you call, configure, or monitor model router.

  • November 2026: Auto (model router without a deployment) is in preview. Set model to auto to call model router with nothing to deploy.
  • November 2026: Model router supports the Responses API natively and adds the Messages API (preview). Per-request routing controls and adaptive session affinity are in preview.
  • November 2026: The Monitor tab and evaluation in the Foundry portal show model router's cost next to single-model baselines.
  • As of November 11, 2026, model router version 2025-11-18 routes across 43 models.

Important

Features or capabilities marked (preview) in this article are in public preview and are subject to the Supplemental Terms of Use for Microsoft Azure Previews. Previews are provided without a service-level agreement, and Microsoft doesn't recommend them for production workloads. Certain features might not be supported or might have constrained capabilities.

November 2026

Auto (preview)

Auto (model router without a deployment) lets you call model router with nothing to deploy or enable: set model to auto in a request to your Foundry resource or project endpoint. Auto is available in every model router region; see Send your first request with Auto.

Native Responses API support

Model router now supports the Responses API on both the resource endpoint and the project endpoint, including streaming, tools, structured outputs, background mode, and multi-turn conversations with previous_response_id or conversation. See Use the Responses API.

Messages API support (preview)

Call model router in the Anthropic Messages format at https://<resource>.services.ai.azure.com/anthropic/v1/messages, where the Claude models deployed in your Foundry resource answer. The Messages API also works with Claude Code and the Claude Agent SDK; see Use the Messages API.

Per-request routing controls (preview)

Send the Foundry-Features: ModelRouterControls=V1Preview header with a routing_config object to set the routing mode, limit the models, or require support for every parameter for a single request. See Control model router routing per request.

Adaptive session affinity, now also for the Responses API (preview)

With the Foundry-Features: ModelRouterControls=V1Preview header, session affinity is now on by default with the new adaptive strategy, even when you don't send routing_config. Adaptive affinity keeps a conversation on the previous model while that model's prompt cache is likely still warm and its quality is close to the best-ranked model. Session affinity now works with the Responses API too. Responses report the strategy in session_affinity.strategy instead of mode. To turn affinity off for a request, set the strategy to none. See Keep a conversation on one model.

Your provisioned and pay-as-you-go deployments serve routed requests first (preview)

When model router selects a model and your Foundry resource has your own deployment of that exact model name and version, that deployment serves the request first: a provisioned throughput (PTU) deployment (Global Provisioned, Data Zone Provisioned, or Regional Provisioned) for OpenAI models, and then a pay-as-you-go deployment (Standard, Global Standard, or Data Zone Standard) for any pool model. Otherwise, router-managed capacity does. Routed traffic fills the capacity that you already pay for first, and there's nothing to configure; see Use your own deployments for routed requests.

Monitor tab for model router deployments

The Monitor tab on your model router deployment, released November 5, 2026, shows requests by model, latency, estimated cost compared with single-model baselines, cache hit rate, and fallback rate. See Use the Monitor tab.

Evaluate model router in the Foundry portal

Choose your model router deployment as the target of a model evaluation, and compare it with a single-model deployment on quality, P50 and P95 latency, and estimated cost, with a breakdown by the models that model router selected. See Evaluate in the Foundry portal.

Eight new models added

Since the October 2026 update, model router version 2025-11-18 adds eight models, for a pool of 43 models as of November 11, 2026. Six of them joined in November 2026, and gpt-6-sol and gpt-6-luna joined on September 29, 2026. o4-mini leaves the pool when it retires on November 19, 2026. To use the new Claude models, deploy them in your Foundry resource first; see Supported models.

  • OpenAI, new in November: gpt-6.1-terra version 2026-10-26, gpt-6.1-sol version 2026-09-29, and gpt-5.1 version 2025-11-13.
  • Anthropic, new in November: claude-opus-5-5 and claude-haiku-5-5, each at version 1.
  • DeepSeek, new in November: DeepSeek-V4.1-Flash version 2026-09-10.
  • OpenAI, added September 29: gpt-6-sol and gpt-6-luna, each at version 2026-09-22.

Automatic model selection in AI Gateway (preview)

An AI Gateway model alias with automatic model selection uses model router to rank the models behind the alias for each request, and the gateway applies your priorities, weights, and policies. There's no model router routing fee during the preview. See Use model router with AI Gateway.

Reorganized model router documentation

Model router documentation is now organized by task, from your first request to monitoring and troubleshooting; start with What is model router?.

October 2026

Eight new models added

Model router version 2025-11-18 now supports 35 models. The routing pool adds the following models:

  • OpenAI: gpt-6-astra version 2026-09-03.
  • Anthropic: claude-fable-5-1, claude-opus-5, and claude-sonnet-5, each at version 1.
  • xAI: grok-4.6 version 1.
  • Fireworks: FW-GLM-5.3, FW-GLM-5.3-Flash, and FW-Kimi-K3, each at version 1.

To use the new Anthropic models with model router, first deploy them to your Foundry resource. For the complete routing pool and deployment requirements, see Supported models.

September 2026

Session affinity for Chat Completions (preview)

Chat Completions applications can now supply an opaque session ID to ask model router to attempt the same eligible model across related conversation turns. Session affinity preserves existing eligibility and fallback behavior and reports whether a request initialized, retained, or switched the model association. For setup and verification guidance, see Keep a session on one model with a session ID.

Per-request routing metadata (preview)

Chat Completions responses from model router identify the serving model and can now include preview observability metadata when the client opts in to the preview response contract. The metadata reports the routing mode, routing-trace latency, ordered model attempts, HTTP status, and returned errors. Multiple attempts can show when automatic fallback occurred for an individual request. For setup and parsing guidance, see Get per-request routing details.

Model router expands to 32 regions

Model router is now available in 32 Azure regions. The four additional regions are Canada Central, North Europe, Norway East, and UAE North. North Europe and Norway East support both Global Standard and Data Zone Standard deployments. Canada Central and UAE North support Global Standard deployments only. For the full list, see Supported regions.

August 2026

Model router expands to 28 regions

Model router is now available in 28 Azure regions. Global Standard was available in all 28 regions, while Data Zone Standard was available in 21. For the full list, see Supported regions.

Model refresh adds GPT-5.6 family and Claude Opus 4.8

Model router version 2025-11-18 now supports 27 models. The routing pool adds gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna version 2026-07-09, and claude-opus-4-8 version 1. To use claude-opus-4-8 with model router, first deploy it to your Foundry resource.

The routing pool removes the retired gpt-5-chat, gpt-5.2-chat, gpt-5.3-chat, DeepSeek-V3.1, and claude-opus-4-1 models.

Configuring a custom model subset is optional. If you previously configured a subset that contains any of these removed models, replace them with models from the current supported models list. If you use the default routing pool, no action is required.

Agentic routing expands to more model providers

For agentic requests, model router can now select eligible open-source (OSS) and Anthropic models in addition to OpenAI models. Model and tool compatibility determine which models are eligible for each request. For more information, see Use model router with agents, agent frameworks, and coding tools.

July 2026

Regional expansion

Model router is now available in three additional regions: Australia East, South India, and West US 3, with both Global Standard and Data Zone Standard deployment types. For the full list of supported regions, see Supported regions.

May 2026

Seven new models added

Model router now supports 28 models. Version 2025-11-18 adds support for seven new models: gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.3-chat, gpt-5.5, claude-opus-4-7, and grok-4-1-fast-reasoning. To use claude-opus-4-7 in your model router deployment, you need to first deploy it to your Foundry resource (see Deploy and use Claude models). grok-4-1-fast-reasoning model router support is in preview.

Increased quota limits

Model router quota limits increased across Tier 1 through Tier 6 for both GlobalStandard and DataZoneStandard deployment types. For the updated limits, see Quota by usage tier.

March 2026

Four new models added

Version 2025-11-18 of model router adds support for four new models: gpt-5.2, gpt-5.2-chat, DeepSeek-V3.2, and claude-opus-4-6. To use claude-opus-4-6 in your model router deployment, you need to first deploy it to your Foundry resource (see Deploy and use Claude models). DeepSeek-V3.2 and claude-opus-4-6 model router support is in preview.

Automatic failover

Model router now includes built-in automatic failover. When a routed model experiences endpoint instability, model router transparently redirects the request to the next most appropriate model, so transient issues with any single model don't disrupt your application. Failover requires no configuration. For more information on the failover feature and how it works with routing mode and model subsets, see Failover.

November 2025

Anthropic models added

Version 2025-11-18 of model router adds support for three Anthropic models: claude-haiku-4-5, claude-opus-4-1, and claude-sonnet-4-5. To include these in your model router deployment, you need to first deploy them yourself to your Foundry resource (see Deploy and use Claude models).

Model router GA version

A new model router model is now available. Version 2025-11-18 includes support for all underlying models in previous versions, as well as 10 new language models.

It also includes new features that make it more versatile and effective.

  • Routing profiles (now called routing modes) let you skew model router's choices to optimize for quality or cost while maintaining a baseline level of performance.
  • Model router supports custom subsets: you can specify which underlying models to include in routing decisions. This gives you more control over cost, compliance, and performance characteristics.
  • Model router supports Global Standard and Data Zone Standard deployment types.

For more information on model router and its capabilities, see the Model router concepts guide.

August 2025

New version of model router (preview)

  • Model router now supports GPT-5 series models.

  • Model router for Microsoft Foundry is a deployable AI chat model that automatically selects the best underlying chat model to respond to a given prompt. For more information on how model router works and its advantages and limitations, see the Model router concepts guide. To use model router with the Chat Completions API, follow the How-to guide.

May 2025

Model router (preview)

Model router for Foundry is a deployable AI chat model that automatically selects the best underlying chat model to respond to a given prompt. For more information on how model router works and its advantages and limitations, see the Model router concepts guide. To use model router with the Chat Completions API, follow the How-to guide.