API Status

Deprecated LLM Models — What to Use Instead

Migration guide with benchmark data for recommended replacements

As LLM providers release newer models, older versions are deprecated and eventually shut down. This guide maps the most common deprecated models to their recommended replacements, with links to current benchmark data for each option. Last updated: 2026-06-11.


Deprecated Model → Replacement Guide

Deprecated ModelProviderStatusRecommended ReplacementNotes
gpt-4-0125-previewOpenAI
deprecated
gpt-4oGPT-4 Turbo preview → GPT-4o. Faster, cheaper, multimodal.
gpt-3.5-turbo-0125OpenAI
deprecated
gpt-4o-miniGPT-3.5 Turbo → GPT-4o Mini. Similar price tier, better quality.
gpt-4-turboOpenAI
deprecated
gpt-4oGPT-4 Turbo → GPT-4o. GPT-4o is faster and cheaper.
text-davinci-003OpenAI
deprecated
gpt-4o-miniLegacy Davinci completion model. Migrate to chat completions API.
claude-2.0Anthropic
deprecated
claude-3-5-haikuClaude 2 → Claude 3.5 Haiku. Similar price, dramatically better performance.
claude-instant-1.2Anthropic
deprecated
claude-3-5-haikuClaude Instant → Claude 3.5 Haiku. Much better quality at comparable cost.
llama3-70b-8192 (Groq)Groq
disabled
llama-3.3-70bLlama 3 70B → Llama 3.3 70B. Drop-in replacement with better benchmarks.

Migration Tips

Test before switching

Run your evaluation set on the new model before cutting over. Newer models often have different output style, token counting, and function-calling behavior.

Update your model string

Most migrations only require changing the model parameter in your API call. For OpenAI migrations, also check that you are using the Chat Completions API (not the legacy Completions endpoint).

Context window changes

Replacement models often have larger context windows. If you were truncating inputs to fit the old model's limit, you may be able to remove that logic.

Compare benchmark data

Check the live benchmarks page to compare throughput and latency between the old and new model before committing.

Frequently Asked Questions

Migrate to gpt-4o. It is faster, cheaper, and multimodal. If you were using GPT-4 Turbo preview for cost efficiency, gpt-4o-mini is even cheaper and handles most tasks well.

GPT-4o Mini (gpt-4o-mini) is the recommended replacement. It is comparable in price to GPT-3.5 Turbo and significantly better at instruction-following and reasoning.

Claude 3.5 Haiku is Anthropic's recommended replacement for Claude 2 and Claude Instant. It offers much better performance at similar pricing.

Check the model's page on this site — deprecated models show a warning banner. You can also check the provider's official deprecation schedule in their documentation.

Providers typically give a deprecation window (often 3–6 months) before shutting down endpoints. After the shutdown date, API calls to the deprecated model will fail. Always migrate before the stated end-of-life date.