Azure GPT, AWS & Vertex Claude on One Key: Channel Model IDs

Update (2026-11-10): as of the 2026-11-10 catalog, the seven vertex/claude-* ids in the table below are not listed, so their rows link no model page and the Vertex comparison page below is not available; the table's other 14 ids are listed. The model catalog and each model page show which ids are on sale.

On 2026-09-22 the Router One catalog lists 21 model ids that carry a channel prefix instead of a vendor prefix: five azure/gpt-* ids, nine aws/claude-* ids and seven vertex/claude-* ids. Each one is the same base model as a default-channel id you may already call — azure/gpt-5.6-sol is GPT-5.6 Sol, aws/claude-sonnet-5 is Claude Sonnet 5 — sold as a separate product: its own routing pool, its own posted rates and tiers, its own quota accounting, and its own line in every trace. The key, base URL, endpoints and request shape do not change; only the model string does.

This guide is what the catalog and the gateway's published behavior let us state about channel ids on that date: which ids exist, what a channel prefix changes and what it leaves alone, which endpoint serves each family, and how to point a client — the OpenAI SDK, Claude Code or Codex CLI — at one. Prices are not printed here; the model catalog and each model page carry the live rate, and the comparison pages listed below render a channel id beside its default twin.

What a channel id is, and what it is not

A channel id names a public sales channel on Router One's catalog. azure/, aws/ and vertex/ are the labels the catalog gave on 2026-09-22 to three channels through which the same base models are served, each with a price list of its own; as of the 2026-10-23 catalog, aistudio/ labels a fourth, the AI Studio channel (see the 2026-10-24 update below). The default channel is everything you called before this change: bare names such as gpt-5.6-sol, and vendor-prefixed names such as openai/gpt-5.6-sol and anthropic/claude-sonnet-5. Those ids keep their existing routing and rates; nothing about them changed on 2026-09-22.

The label is a product identity, not a disclosure. Router One does not publish which suppliers serve any channel, how many candidate routes a channel has, what the routes cost the gateway, or which regions they run in, and the customer trace does not either (routing facts). A channel id therefore does not come with a data-residency, region or compliance statement from Router One. If your own policy needs a specific cloud channel, the channel label and the per-request trace are what the gateway gives you to work with; the rest belongs to your own review.

The 21 ids

Every id below was listed in the 2026-09-22 catalog check. The catalog entry, not this table, is authoritative for context window, input modalities, capability flags and rates — open the model page before you send traffic.

Two things the table shows are easy to miss. The channel lineups are not copies of the default lineup: aws/claude-opus-4.5 and aws/claude-sonnet-4.5 have no default-channel twin in the 2026-11-10 catalog (nor did vertex/claude-sonnet-4.5 in the 2026-09-22 one), and the default channel lists Claude Opus 4.7 only as anthropic/claude-opus-4.7-thinking; as of the 2026-10-15 catalog, Router One no longer lists openai/gpt-5.5, so azure/gpt-5.5 has no default-channel twin either; the AWS channel carries a Haiku 4.5 id, which the Vertex channel's 2026-09-22 Claude lineup did not include; and no channel carries a -thinking id or Claude Fable 5. And azure/gpt-5.6-luna is a new channel product, not the default-channel openai/gpt-5.6-luna id.

Update (2026-10-15): as of the 2026-10-15 catalog, the Vertex channel also lists two Gemini ids that the 2026-09-22 table above does not include: vertex/gemini-3.8-flash (base model Gemini 3.8 Flash, beside google/gemini-3.8-flash) and vertex/gemini-4-argon (base model Gemini 4 Argon, beside google/gemini-4-argon). The channel rules below apply to them as well; they are served on POST /v1/chat/completions only, and no plan tier lists either in the 2026-10-15 plan response.

Update (2026-10-17): as of the 2026-10-17 catalog, the Azure channel also lists two GPT ids that the 2026-09-22 table above does not include: azure/gpt-6-sol (base model GPT-6 Sol, beside openai/gpt-6-sol) and azure/gpt-6.1-sol (base model GPT-6.1 Sol, beside openai/gpt-6.1-sol). The channel rules below apply to them as well; like the other azure/gpt-* ids they are served on POST /v1/chat/completions and natively on POST /v1/responses. No plan tier lists either in the 2026-10-17 plan response, so their calls bill the wallet at their posted rates. In the same response, openai/gpt-6-sol and openai/gpt-6.1-sol are in the Premium models tier of the Pro, Max and Ultra plans.

Update (2026-10-24): as of the 2026-10-23 catalog, the AI Studio channel lists six Gemini ids that the 2026-09-22 table above does not include: aistudio/gemini-3.5-flash (base model Gemini 3.5 Flash, beside google/gemini-3.5-flash), aistudio/gemini-3.5-flash-lite (base model Gemini 3.5 Flash Lite, beside google/gemini-3.5-flash-lite), aistudio/gemini-3.6-flash (base model Gemini 3.6 Flash, beside google/gemini-3.6-flash), aistudio/gemini-3.7-flash (base model Gemini 3.7 Flash, beside google/gemini-3.7-flash), aistudio/gemini-3.8-flash (base model Gemini 3.8 Flash, beside google/gemini-3.8-flash) and aistudio/gemini-4-argon (base model Gemini 4 Argon, beside google/gemini-4-argon). The channel rules below apply to them as well; they are served on POST /v1/chat/completions only. No plan tier lists an aistudio/ id in the 2026-10-24 plan response, so their calls bill the wallet at their posted rates.

Update (2026-10-31): as of the 2026-10-30 catalog, the Azure channel also lists a GPT id that the 2026-09-22 table above does not include: azure/gpt-6-luna (base model GPT-6 Luna, beside openai/gpt-6-luna). The channel rules below apply to it as well (for Codex CLI, the GPT-6 Luna API guide gives the model line to use); like the other azure/gpt-* ids it is served on POST /v1/chat/completions and natively on POST /v1/responses (observed 2026-10-31), and its model page lists its own posted rates. It is in no plan tier as of the 2026-10-31 plan response, so its calls bill the wallet per token.

What changes when you pick a channel

The gateway treats a channel id as a product in its own right. Six consequences follow, and they are the reason to choose — or not to choose — a channel id.

1. Routing stays inside the channel. A request for azure/gpt-5.6-sol is served only by Azure-channel routes for that model. Retries after a retryable upstream failure stay inside the channel; the request never falls back to the default channel, and automatic routing cannot move it there. When no channel route is available, the request fails with the no-eligible-route error the error codes page describes rather than being answered by another channel.
2. The default id never borrows channel capacity. openai/gpt-5.6-sol keeps the fallback scope it had; the fact that a channel serves the same base model does not add that channel's routes to the default id's candidates.
3. Rates are the channel's own. Every rate line the model page shows — input, output, and the whole-request long-context tier the Azure GPT ids carry above 272,000 input tokens — is posted per channel id and does not inherit the default id's promotional discount. On 2026-09-22 the default GPT ids are where the up to 90% off official prices positioning applies; the Azure ids carry a smaller discount off their own list rate, which can differ from the default id's; and the AWS and Vertex Claude ids bill at their posted list rate with no promotional discount. Read the current figures on the model page; they change.
4. Quota and discounts follow the channel id. A subscription quota that covers a default id does not automatically cover its channel twin: unless the live plan response lists the channel id in a model tier, calls to it bill to wallet balance at the posted rates, and the cost calculator reports that no published plan quota covers the model. Account-level discounts are likewise applied to the channel id you called, not to the base model name.
5. The channel survives into the response and the ledger. The model field in the response, the row in Dashboard → Logs and the settled charge carry the channel id you sent, so channel traffic can be filtered, budgeted with a dedicated key and its maxSpend, and reconciled by request_id like any other model (per-request observability).
6. The catalog shows the channel and the customer price, nothing behind them. Supplier names, candidate counts and procurement prices are not published for any channel, the default channel included.

Which endpoint serves a channel id

Channel ids follow their base family's endpoint rule, which the API compatibility facts state and each model page lists:

• azure/gpt-* ids are GPT-family: POST /v1/chat/completions, and POST /v1/responses served natively — the path Codex CLI speaks.
• aws/claude-* ids are Claude-family: POST /v1/chat/completions, and POST /v1/messages in the Anthropic wire format — the path Claude Code speaks.
• vertex/gemini-* ids (listed as of the 2026-10-15 catalog) are Gemini-family: POST /v1/chat/completions only.
• aistudio/gemini-* ids (listed as of the 2026-10-23 catalog) are Gemini-family: POST /v1/chat/completions only.

An id sent to an endpoint that does not serve it — a Claude channel id on /v1/responses, for instance — is rejected with HTTP 400 invalid_request_error before any model is called, and the message names the path to use. No channel id ends in -thinking, so the thinking default described on the compatibility sheet does not apply to them.

There is no channel field in the request body. The model id is the only selector: azure/gpt-5.6-sol selects the Azure channel, gpt-5.6-sol and openai/gpt-5.6-sol select the default channel, and no header or extra parameter changes that.

Calling one

Chat Completions with an Azure GPT id — the request you send today, with a different model string:

curl https://api.router.one/v1/chat/completions \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "azure/gpt-5.6-sol",
    "messages": [{"role": "user", "content": "Explain what a sales channel is on an API catalog, in one sentence."}]
  }'

Messages with an AWS Claude id:

curl https://api.router.one/v1/messages \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "aws/claude-sonnet-5",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Explain what a sales channel is on an API catalog, in one sentence."}]
  }'

Claude Code: keep ANTHROPIC_BASE_URL=https://api.router.one and ANTHROPIC_AUTH_TOKEN as the Claude Code setup describes, and set ANTHROPIC_MODEL=aws/claude-sonnet-5 (or another aws/claude-* id). Claude Code sends the id unchanged on /v1/messages.

Codex CLI: with the config the Codex and Responses API guide shows, set the top-level model = "azure/gpt-5.6-sol" (not inside the [model_providers] table). Codex sends the id unchanged on /v1/responses, where the Azure GPT ids are served natively.

Then open Dashboard → Logs: each call is one row with the channel id as its model, tokens, the wallet charge or plan quota it used, latency and status. If you run a channel and its default twin side by side, give each its own key with a maxSpend cap so the comparison has a budget (per-key cost tracking).

Compare a channel with its default twin

Comparison pages render a channel id beside its default-channel twin from the live catalog — the price gap on a 1M-input + 1M-output mix, the context window and the endpoint family, refreshed within the hour:

• GPT-5.6 Sol (Azure) vs GPT-5.6 Sol
• Claude Sonnet 5 (AWS) vs Claude Sonnet 5
• Claude Opus 5 (Vertex AI) vs Claude Opus 5 — not available as of the 2026-11-10 catalog, which does not list vertex/claude-opus-5

These pages state price facts only; whether a channel behaves differently for your prompts is a question for your own A/B run.

When a channel id is the right call

• You need traffic pinned to one channel. Latency, availability and error observations in your traces are then attributable to that channel alone, because no request was quietly served elsewhere.
• Your procurement or review policy names a cloud channel. The catalog label and the trace give you a per-request record of which channel product answered; Router One supplies no certification or residency statement on top of that.
• You are evaluating the channel itself. Run the channel id and the default id under separate keys and compare traces before moving volume.

When cost decides, the default ids were the answer for the 21 ids listed on 2026-09-22: they carried the deeper discount on the same base models; for a channel id listed later, compare its model page with its default twin's. A channel id is a deliberate choice to pay the channel's rate for the channel's isolation.

What is not offered

• No request-body channel selector; model is the only way to choose a channel.
• No promise that a channel lineup stays complete or stable. Ids are listed and retired on the catalog's schedule, and the comparison pages above return 404 rather than wrong numbers if a side leaves.
• No -thinking channel ids and no Claude Fable 5 channel id as of 2026-09-22.
• No supplier, region, candidate-count or procurement-price disclosure, for channel or default ids.

FAQ

Does azure/gpt-5.6-sol give me the same model as openai/gpt-5.6-sol? The same base model, sold as a separate product: its own routing pool, its own posted rates, its own quota accounting. The two catalog entries may also differ in published details, so read both model pages rather than assuming parity.

Will a channel id fall back to the default channel when the channel is busy? No. Retries stay inside the channel; when no channel route is available the request fails with the no-eligible-route error rather than being served by another channel. The default id, in turn, never uses channel routes.

Does my Pro, Max or Ultra plan cover the channel ids? Only if the live plan response lists the channel id in a model tier. A quota that covers openai/gpt-5.6-sol does not cover azure/gpt-5.6-sol; without listed coverage, channel calls bill to wallet balance at the posted rates, and the cost calculator says so for the model you pick.

Are the channel ids cheaper? Not on 2026-09-22, for the 21 ids listed that day: the default ids carried the deeper promotional discount; the Azure ids carried a smaller discount, and the AWS and Vertex Claude ids billed at their posted list rate. A channel id listed later has its own rates, so compare its model page with its default twin's; the comparison pages render the gap live, and this post prints no rates.

Which endpoint do Claude Code and Codex use with a channel id? Claude Code sends aws/claude-* ids on /v1/messages; Codex sends azure/gpt-* ids on /v1/responses, where the Azure GPT ids are served natively. Both are the paths their default-channel twins already use.

Next steps

• Pick a channel id on the model catalog and read its endpoints, tiers and live rate; the discount qualifier is on the pricing page.
• Read the pricing facts and routing facts for the channel rules in fact-sheet form.
• For the rest of the September lineup, including Grok 4.7, see the September 2026 new-model guide.