The newest OpenAI name in Microsoft's Azure model registry is gpt-6.1-terra, dated October 26, 2026[1]. It joins a registry that also carries gpt-6.1-sol from September 29, the three GPT-6 models gpt-6-astra, gpt-6-luna and gpt-6-sol from September 3 and 22, and the GPT-5.6 trio gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna from July 9[1]. Three GPT generations in under four months, reusing the same small set of codenames each time. An OpenAI model name therefore reads on three layers: the version number names the generation, the codename names the tier inside that generation, and the suffix names size or purpose. Read any single layer alone and the name points to the wrong model. The checkable record sits in platform documentation: Microsoft's Foundry pages register IDs with dates and lifecycle status[1][2], and AWS's Bedrock documentation lists the same codenames as Generally Available under its own prefixes[3].
The three layers of a model name
The version number moves fastest. Between November 13, 2025 and October 26, 2026 the registry added 5.1, 5.2, 5.3 (as gpt-5.3-codex), 5.4, 5.5, 5.6, 6 and 6.1 — eight version numbers in under twelve months[1]. The number tracks the generation, and the registry treats each version-plus-codename pairing as a separate model: gpt-5.6-sol and gpt-6.1-sol share the codename Sol, registered July 9 and September 29 as two distinct IDs[1].
The codename marks the tier within a generation. GPT-6 consists of Astra, Luna and Sol; GPT-5.6 of Sol, Terra and Luna; GPT-6.1 so far of Sol and Terra[1]. The same codename reappears across generations: Sol labels one model in 5.6, a different one in 6, a third in 6.1; Terra spans 5.6 and 6.1; Luna spans 5.6 and 6[1]. A codename on its own can therefore return up to three different models on this page — only the full ID, version plus codename plus suffix, resolves to one[1]. Astra appears once, in GPT-6, and carries its own safety note in the registry: "In certain circumstances, Astra may apply enhanced safety controls when safety systems identify elevated risk"[1]. The registry fixes which IDs exist and when; it attaches no positioning definitions to the codenames.
The suffix covers size and purpose. -mini and -nano mark smaller variants; in the 5.4 generation, gpt-5.4-mini and gpt-5.4-nano are registered at 400,000 tokens against the base model's 1,050,000[1]. -codex marks the models registered for Codex CLI and the Codex VS Code extension (gpt-5.1-codex, gpt-5.2-codex, gpt-5.3-codex), and -max the added reasoning setting on gpt-5.1-codex-max, which supports reasoning_effort at xhigh while plain gpt-5.1 defaults reasoning_effort to none[1]. Within the GPT-5 line, -pro marks two models, gpt-5.4-pro and gpt-5-pro, both registered as Responses API without Chat Completions — a trait they share with the -codex line[1]. -chat marks the Chat Completions snapshot variants, and those are exactly the names already retired. The image models follow the same pattern: gpt-image-1, gpt-image-1-mini, gpt-image-1.5, gpt-image-2, gpt-image-2.5-flare, gpt-image-2.5-sunburst[1].
One registered name fits no layer. gpt-chat-latest has no codename, carries a Preview tag, and rolls forward through dated snapshots — May 5, May 28, and on to October 7, 2026[1]. The same model is spelled two ways across platforms: "In the OpenAI API, this model is referred to as chat-latest. In Microsoft Foundry, the product name is gpt-chat-latest"[1].
What Azure registers, October 31, 2026
The page sets its own scope: it lists "a selection of" Foundry Models sold by Azure, "excluding deprecated and retired models"[1]. The scope line and the tables are not perfectly aligned — the same page lists gpt-4o versions that the retirement schedule marks Deprecated[2]. What it registers under the OpenAI-side naming includes[1]:
| Series | Model IDs and registration dates |
|---|---|
| GPT-6.1 | gpt-6.1-sol (Sep 29), gpt-6.1-terra (Oct 26) |
| GPT-6 | gpt-6-astra (Sep 3), gpt-6-luna (Sep 22), gpt-6-sol (Sep 22) |
| GPT-chat-latest (Preview) | gpt-chat-latest, snapshots May 5 through Oct 7, 2026 |
| GPT-5.6 | gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna (all Jul 9) |
| GPT-5.5 | gpt-5.5 (Apr 24) |
| GPT-5.4 | gpt-5.4, gpt-5.4-pro (Mar 5); gpt-5.4-mini, gpt-5.4-nano (Mar 17) |
| GPT-5.3 | gpt-5.3-codex (Feb 24) |
| GPT-5.2 | gpt-5.2 (Dec 11, 2025), gpt-5.2-codex (Jan 14) |
| GPT-5.1 | gpt-5.1, gpt-5.1-codex, gpt-5.1-codex-mini (Nov 13, 2025), gpt-5.1-codex-max (Dec 4, 2025) |
| GPT-5 | gpt-5, gpt-5-mini, gpt-5-nano (Aug 7, 2025), gpt-5-codex (Sep 11), gpt-5-pro (Oct 6) |
| o-series | o1, o1-mini, o3, o3-mini, o3-pro, o4-mini, still listed; the page also mentions o3-deep-research in prose |
| codex-mini | codex-mini, a fine-tuned version of o4-mini |
| GPT-4.1 | gpt-4.1, gpt-4.1-mini, gpt-4.1-nano |
| GPT-4o | gpt-4o and gpt-4o-mini versions |
| GPT-4 Turbo | gpt-4 (turbo-2024-04-09) |
| gpt-oss | open-weight reasoning models |
| Image | gpt-image-1, gpt-image-1-mini, gpt-image-1.5, gpt-image-2, gpt-image-2.5-flare, gpt-image-2.5-sunburst |
| Video | sora-2 |
| computer use | computer-use-preview (Mar 11, 2025), an experimental model for the Responses API computer use tool |
| Embeddings | text-embedding-ada-002 (versions 1 and 2), text-embedding-3-large, text-embedding-3-small |
| Audio | named in the page's highlights; detailed table not reproduced here |
The list is partial: the page's highlights name an audio series whose model table is absent from the copy this article draws on[1]. Three registered details matter for anyone calling these models. GPT-6 deployments "use separate pricing categories for short-context and long-context requests", and "Each GPT-6 model handles both request types", so one ID can be billed two ways depending on request length[1]. On GPT-5.6, the Chat Completions API and function tools cannot run together "unless reasoning_effort is none", which moves tool calling to the Responses API[1]. The GPT-6.1 pair is registered at 1,050,000 tokens of context, split 922,000 input and 128,000 output, the three GPT-5.6 models carry the same total, and the 5.4 -mini/-nano rows sit at 400,000[1].
The same names on AWS Bedrock
The codenames are registered outside Azure as well. Amazon Bedrock's documentation lists openai.gpt-6.1-sol, openai.gpt-6-astra, openai.gpt-5.6-sol, openai.gpt-5.6-terra and openai.gpt-5.6-luna as Generally Available[3], and it draws the generation line at the same place Azure's dates do: "GPT-5.6 and later models also support Explicit Prompt Caching", while for models before GPT-5.6, "such as openai.gpt-5.5 and openai.gpt-5.4", implicit caching is automatic with no parameters[3]. Bedrock's example request targets global.openai.gpt-5.6-sol — the system inference profile, the same model ID prefixed with a global. scope[3]. The platform prefix changes and the codename stays, so a comparison across platforms holds only when it names the version and the codename together.
Names already retired and their listed replacements
Microsoft's retirement schedule records every exit with a lifecycle stage, a date and a suggested replacement, and states that the Replacement column "identifies Microsoft's recommended replacement and, where supported, the target for automatic upgrades at retirement"[2]. The schedule tracks full IDs, not codenames — it records which IDs have left the list and which are scheduled to. Four -chat names, across six version rows, are already gone: gpt-5-chat (versions 2025-08-07 and 2025-10-03), gpt-5.1-chat, gpt-5.2-chat (versions 2025-12-11 and 2026-02-10) and gpt-5.3-chat, retired May 13 and June 29, 2026. All six rows list the same replacement: gpt-chat-latest[2]. The retired traffic funnels into the one name with no codename, which is still receiving new snapshots[1].
| Retired ID (version) | Retired on | Listed replacement |
|---|---|---|
gpt-5-chat (2025-08-07) |
Jun 29, 2026 | gpt-chat-latest |
gpt-5-chat (2025-10-03) |
May 13, 2026 | gpt-chat-latest |
gpt-5.1-chat (2025-11-13) |
Jun 29, 2026 | gpt-chat-latest |
gpt-5.2-chat (2025-12-11) |
May 13, 2026 | gpt-chat-latest |
gpt-5.2-chat (2026-02-10) |
Jun 29, 2026 | gpt-chat-latest |
gpt-5.3-chat (2026-03-03) |
Jun 29, 2026 | gpt-chat-latest |
The older reasoning line has one confirmed exit: o1-preview, retired July 28, 2025[1]. Exits already scheduled include gpt-4o (2024-05-13) on December 9, 2026, with gpt-5.6-sol as the listed replacement, while the two later gpt-4o versions point to gpt-5.1; codex-mini on November 15, 2026; the gpt-4.1 family on April 14, 2027; and, from there, gpt-5 on February 9, 2027, gpt-5.1 on May 15, gpt-5.2 on June 8, gpt-5.3-codex on August 24, gpt-5.4 on September 2, gpt-5.5 on October 26, and the GPT-5.6 trio on January 11, 2028[2]. Even names registered this July carry an end date.
What the rule does with the October names
gpt-6.1-terra reads as generation 6.1, Terra tier, no suffix — the second model registered in the 6.1 series, 27 days after gpt-6.1-sol[1]. The October 7 gpt-chat-latest snapshot is the rolling alias doing its job: the retirement table sends every dead -chat name to it[1][2]. For anyone matching model names across Azure, Bedrock and the OpenAI API, the working form is the full ID — version number for the generation, codename for the tier inside it, suffix for size or purpose — checked against Microsoft's Foundry model list and retirement schedule, or AWS's Bedrock model tables, and never against a codename on its own.