Anthropic retires Claude Sonnet 4 and Opus 4
Anthropic will retire Claude Sonnet 4 and Claude Opus 4 from native support on the Claude API on June 15 2026 at 9AM PT. Users must migrate to newer models including Opus 4.7. The older Opus 4 had already shown reduced availability and higher error rates. The move follows the prior unannounced removal of Opus 4 from Claude.ai. AI safety researcher Neil Chowdhury noted that the Sonnet 4 retirement ends access to a model free of constitutional classifiers applied in newer versions.
#1174Maksym Andriushchenko@MAKSYM_ANDRi actually missed this: both OpenAI and Anthropic seem to be winding down their fine-tuning APIs. the most recent models like GPT-5.x and Claude 4.x are not available at all. why? misuse risks?
this also seems relevant for continual learning (e.g., RL on user rollouts from Claude Code / Codex CLI) that produces personalized LoRA adapters for different users. probably the main reason it hasn't been deployed is that it's a potential nightmare for safety and alignment.
for example, this is what https://developers.openai.com/api/docs/guides/supervised-fine-tuning says:
Maksym Andriushchenko@maksym_andri actually missed this: both OpenAI and Anthropic seem to be winding down their fine-tuning APIs. the most recent models like GPT-5.x and Claude 4.x are not available at all. why? misuse risks? this also seems relevant for continual learning (e.g., RL on user rollouts from Claude Code / Codex CLI) that produces personalized LoRA adapters for different users. probably the main reason it hasn't been deployed is that it's a potential nightmare for safety and alignment.
@PandaAshwinee hmm, then why are they getting rid of it? is it hard infra-wise? or compute shortage?
Ashwinee Panda@PandaAshwinee@maksym_andr no way they’re getting rid of finetuning bc it’s too risky but you can still just multi turn prefill the model right?
#1323Neil Chowdhury@CHOWDHURYNEILthis model deprecation will probably stifle my work + other safety researchers’. Sonnet 4 is the best Anthropic model not subject to constitutional classifiers, which frequently over-fire on model behavior research, even if it has nothing to do with chem/bio/jailbreaks
@maksym_andr I guess yet another casualty of compute shortage
Maksym Andriushchenko@maksym_andri actually missed this: both OpenAI and Anthropic seem to be winding down their fine-tuning APIs. the most recent models like GPT-5.x and Claude 4.x are not available at all. why? misuse risks? this also seems relevant for continual learning (e.g., RL on user rollouts from Claude Code / Codex CLI) that produces personalized LoRA adapters for different users. probably the main reason it hasn't been deployed is that it's a potential nightmare for safety and alignment.
@maksym_andr no way they’re getting rid of finetuning bc it’s too risky but you can still just multi turn prefill the model right?
Maksym Andriushchenko@maksym_andri actually missed this: both OpenAI and Anthropic seem to be winding down their fine-tuning APIs. the most recent models like GPT-5.x and Claude 4.x are not available at all. why? misuse risks? this also seems relevant for continual learning (e.g., RL on user rollouts from Claude Code / Codex CLI) that produces personalized LoRA adapters for different users. probably the main reason it hasn't been deployed is that it's a potential nightmare for safety and alignment.