AI glossary · Models and context
What is a model snapshot (vs an alias)?
Also called: pinned model version, dated model ID, model alias
Definition
A model snapshot is a fixed version of a model behind a specific API model ID, so requests to that ID keep getting the same model until it is retired, unlike an alias that the provider can move.
Explained
How it works
Providers name models in two ways. An alias is a convenient name that points to one version and can later be pointed at another. A snapshot ID is pinned: OpenAI says snapshots “lock in a specific version of the model so that performance and behavior remain consistent”.
Each provider does it differently. At OpenAI, a model name is an alias for a default snapshot, and dated IDs are the snapshots. Anthropic says every Claude model ID is pinned; from the 4.6 generation on the IDs carry no date, and only older models have aliases. Google calls codes such as gemini-3.8-flash stable, gives preview codes at least two weeks’ notice before deprecation, and hot-swaps -latest aliases with each new release.
Pinned is not forever. Every ID has its own retirement date, and Anthropic notes that serving changes around a fixed model, such as routing, safety classifiers and sampling, can still shift behaviour slightly.
Example
Aliases and snapshots in real model IDs
The table shows IDs from our model ID reference, checked against each provider’s docs. Look at gpt-4o: it points to gpt-4o-2024-08-06, not to the newer gpt-4o-2024-11-20. An alias means “whichever version the provider chose”, not “the latest”.
Note too that claude-opus-5-5 has no date but is not an alias, while gpt-5.5 looks just as plain and is one. You can’t tell from the string: check the provider’s page, or the Claude, OpenAI and Gemini lists.
| Model | ID | Kind | What it means |
|---|---|---|---|
| GPT-5.5 | gpt-5.5 | Alias | Points to the default snapshot, gpt-5.5-2026-04-23 |
| GPT-5.5 | gpt-5.5-2026-04-23 | Snapshot | Never changes |
| GPT-4o | gpt-4o | Alias | Points to gpt-4o-2024-08-06 |
| Claude Haiku 4.5 | claude-haiku-4-5 | Alias | Points to claude-haiku-4-5-20251001 |
| Claude Opus 5.5 | claude-opus-5-5 | Snapshot (no date) | Never changes |
| Gemini 3.8 Flash | gemini-3.8-flash | Stable code | Google says stable models usually don’t change |
From our model ID reference, checked against each provider’s documentation on 2026-10-11.
Cost and quality
Why it matters
In production, pin the snapshot so a provider update can’t silently change outputs that your prompts, parsers and tests depend on. Move to a new snapshot on purpose, after running your evaluations on it.
Watch the retirement dates: a pinned ID stops working when it is retired. Aliases are fine for experiments and for tools where you always want the provider’s current pick.
Don’t mix up
Common confusions
- Model name vs model ID
- “Claude Opus 5.5” is a product name; the API needs the exact ID string,
claude-opus-5-5, with dashes where the name has dots. Copy IDs from the provider’s list rather than typing them from the product name. - Pinned snapshot vs identical outputs
- Pinning removes one source of change, the model version. Sampling still varies from call to call; see temperature.
Go deeper
Try it and read more
- Free toolModel ID ReferenceCopy current API model IDs for Claude, GPT, Gemini and more.
- Free toolOpenAI model IDs: the GPT model names listThe GPT model names list for the OpenAI API: model IDs and dated snapshots, context windows, and the date each deprecated model shuts down.
- Free toolClaude model IDs for the API, Bedrock, Google Cloud and FoundryEvery Claude model ID to copy, from claude-opus-5-5 to dated snapshots and aliases, with Amazon Bedrock, Vertex AI and Foundry IDs and retirement dates.
Related
Related terms
- OpenAI-compatible APIAn OpenAI-compatible API is a model API that accepts OpenAI’s Chat Completions request format, so you can call it with the official OpenAI SDK by changing only the base URL, the API key and the model name.
- TemperatureTemperature is a sampling setting that divides a model’s next-token scores before they become probabilities, so low values make the likeliest token dominate and high values spread the choice across more tokens.
- Context windowA context window is the maximum number of tokens a language model can work with in one request, counting the system prompt, tool definitions, conversation history, documents and the reply it writes.