Model names and versions
A model name usually has three parts: a family (Claude, GPT, Gemini), a generation number, and a size or speed tier; the exact ID you send in code may also pin a dated snapshot. Knowing which kind of ID you use tells you whether the model can change under you, and when it will stop working.
Tiers
Each vendor sells bigger, slower, pricier models and smaller, faster, cheaper ones under its own names (from each vendor's models page, linked on AI providers and models, checked 2026-10-28):
| Vendor | Tier names, most capable to fastest |
|---|---|
| Anthropic | Fable, Opus, Sonnet, Haiku |
| OpenAI | Astra, Sol, Terra, Luna |
| Google Gemini | Pro, Flash, Flash-Lite (where Gemini 4 Argon fits is not stated on the page) |
| Alibaba Qwen | Max, Plus, Flash |
Snapshots and aliases
In code you send a model ID. Some IDs are fixed snapshots; others are aliases that move to newer versions:
- Anthropic: "Starting with the Claude 4.6
generation, model IDs use a dateless format", and such an ID "is not an
alias. It is the snapshot." Older models use dated IDs such as
claude-sonnet-4-5-20250929(Model IDs and versioning). - OpenAI: older snapshots carry a date in the ID,
such as
gpt-4.1-2025-04-14, which OpenAI's deprecations page lists for shutdown on January 14, 2027 (OpenAI deprecations). - Google Gemini: a stable name "Points to a specific stable model. Stable models usually don't change"; preview "may be used for production"; latest points to the newest release and changes over time; experimental models are not for production (Gemini models).
- Mistral: IDs follow "a model-name-major-minor
format, for example mistral-medium-3-5", and
-latestaliases exist, with a warning: "Using an alias may expose you to silent updates in model behavior and pricing" (Mistral model lifecycle). - xAI: a bare model name is "aliased to the latest
stable version",
-latestto the latest version, and a dated name "will not be updated" (xAI models).
All checked 2026-10-28. In production, pin a fixed ID and upgrade on purpose, after testing.
Retirement
Models are retired on a schedule, and calls to a retired ID fail. Published notice periods (checked 2026-10-28):
- Anthropic: "at least 60 days' notice before model retirement for publicly released models" (deprecations).
- Mistral: 6 months for generally available models, 1 month for Labs, preview and third-party models; after retirement, "requests to its identifiers fail with a 404 error" (model lifecycle).
- Google Vertex AI: "gives you at least 45 days to migrate" (model versions).
- OpenAI: "At least 6 months" for generally available models, "At least 3 months" for specialized variants, and possibly "much shorter notice, such as 2 weeks" for previews (deprecations).
Put the model ID in configuration, not scattered through code, and subscribe to your provider's deprecation notices.
Related pages
- AI providers and models: who makes what, and how to choose
- From idea to deployed AI app: changing models safely
See also
Glossary terms used on this page: Model, Provider, Snapshot.
Sources
All checked on 2026-10-28.
- Anthropic: Model IDs and versioning, Deprecations
- OpenAI: Models, Deprecations
- Google: Gemini models, Vertex AI model versions
- Mistral: Model lifecycle
- xAI: Models
First version written with AI assistance and checked against the sources on 2026-10-28. Found a mistake? Fix it or open an issue. Text licensed under CC BY 4.0.