Free · runs in your browser

Is your LLM model deprecated?

Type a model id to check whether it’s retired, deprecated, or legacy — and see the recommended replacement. A silent model shutdown is one of the fastest ways to break an AI feature in production.

Showing all known entries. Type a model id to filter.

text-davinci-003 · OpenAI

Retired

→ Use gpt-4o-mini or gpt-4o

Legacy Completions models are retired; move to the Chat Completions / Responses API.

Also covers: text-davinci-002, davinci, curie, babbage, ada

gpt-3.5-turbo-0301 · OpenAI

Retired

→ Use gpt-4o-mini

Older GPT-3.5 snapshots are retired; gpt-4o-mini is cheaper and stronger.

Also covers: gpt-3.5-turbo-0613, gpt-3.5-turbo-16k

gpt-4-0314 · OpenAI

Deprecated

→ Use gpt-4o

Older GPT-4 snapshots are superseded by GPT-4o (faster, cheaper, 128K context).

Also covers: gpt-4-0613, gpt-4-32k

gpt-4-vision-preview · OpenAI

Deprecated

→ Use gpt-4o

Vision is built into GPT-4o; the standalone vision preview is deprecated.

Also covers: gpt-4-1106-vision-preview

text-embedding-ada-002 · OpenAI

Legacy

→ Use text-embedding-3-small or text-embedding-3-large

The v3 embeddings are cheaper and better; ada-002 is legacy.

claude-1 · Anthropic

Retired

→ Use claude-haiku-3.5 or claude-sonnet-4.5

Claude 1 / Instant are retired.

Also covers: claude-1.3, claude-instant-1, claude-instant-1.2

claude-2 · Anthropic

Retired

→ Use claude-sonnet-4.5

Claude 2.x is retired; Sonnet 4.5 is the balanced successor.

Also covers: claude-2.0, claude-2.1

claude-3-opus · Anthropic

Legacy

→ Use claude-opus-4 / claude-sonnet-4.5 / claude-haiku-3.5

Claude 3 / 3.5 are superseded by the Claude 4 family.

Also covers: claude-3-sonnet, claude-3-haiku, claude-3.5-sonnet

text-bison · Google

Retired

→ Use gemini-2.0-flash or gemini-2.5-pro

PaLM 2 (bison) models are retired; move to Gemini.

Also covers: chat-bison, code-bison, PaLM 2

gemini-1.0-pro · Google

Deprecated

→ Use gemini-2.5-pro

Gemini 1.0 is deprecated; 2.5 Pro is the current frontier tier.

Also covers: gemini-pro, gemini-1.0-pro-vision

gemini-1.5-pro · Google

Legacy

→ Use gemini-2.5-pro or gemini-2.0-flash

Gemini 1.5 is superseded by the 2.x family.

Also covers: gemini-1.5-flash

llama-2 · Meta

Legacy

→ Use llama-3.3-70b

Llama 2 is superseded by Llama 3.x.

Also covers: llama-2-70b, llama-2-13b, codellama

mistral-tiny · Mistral

Legacy

→ Use mistral-large or mistral-small (current)

Early Mistral aliases are superseded by the current lineup.

Also covers: mistral-medium, mistral-small (legacy)

A community guide, not gospel — providers change lifecycles often, so verify against the provider’s own deprecation page. retired = no longer callable · deprecated = scheduled for shutdown · legacy = still works but superseded.

Never ship on a dead model again

VernaOne syncs each provider’s live model catalog and automatically pauses — or falls back — any prompt whose model went inactive, then alerts you. The decommission is caught before your users see it.

See model-sync & alerts →

Frequently asked

Is my LLM model deprecated?

Type the model id above to check it against a guide of well-known deprecated, retired, and legacy models with their recommended replacements — for example, gpt-4-0314, claude-2, gemini-1.5-pro, and text-davinci-003. Providers update lifecycles often, so always confirm on the provider’s own deprecation page.

What is the difference between retired, deprecated, and legacy?

Retired means the model is no longer callable and requests fail. Deprecated means it still works but is scheduled for shutdown — you should migrate. Legacy means it still works but has been superseded by a better or cheaper model.

What happens if I keep calling a deprecated model?

When the provider removes it, every request pinned to that model starts failing — usually a 404 or invalid-model error — and you often find out from an error spike or a customer, not the provider. Migrating early, and having a fallback, prevents the outage.

How do I stop a model shutdown from breaking production?

Don’t rely on manual checks. VernaOne syncs each provider’s live model catalog and runs a health-check that automatically pauses any prompt whose model went inactive (with no active fallback) and alerts you — or falls back to a working model if you configured one. A silent decommission never reaches your users.