ChatGPT

OpenAI to Retire GPT-4o and Legacy Models on Feb 13 — Full Migration to GPT-5.2 Series

OpenAI will retire GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini on February 13, migrating all ChatGPT conversations to the GPT-5.2 series. GPT-5.2 delivers 30% fewer hallucinations, 80% on SWE-Bench Verified, and significant gains across all benchmarks.

OpenAI ChatGPT GPT-5.2 GPT-4o AI Benchmark

End of the Legacy Era

OpenAI announced that the following models will be retired from ChatGPT on February 13, 2026:

  • GPT-4o
  • GPT-4.1 / GPT-4.1 mini
  • o4-mini
  • GPT-5 Instant / GPT-5 Thinking

After retirement, all conversations will automatically migrate to the GPT-5.2 series. The API will not be immediately affected — OpenAI has committed to providing adequate notice before any API deprecation.

GPT-5.2 Series Lineup

GPT-5.2 offers three variants tailored to different use cases:

ModelBest ForAPI Name
GPT-5.2 InstantEveryday queries, search, translationgpt-5.2-chat-latest
GPT-5.2 ThinkingCoding, math, long-document summarizationgpt-5.2
GPT-5.2 ProHigh-difficulty questions, maximum reliabilitygpt-5.2-pro

Key Benchmarks

GPT-5.2 Thinking delivers substantial improvements over its predecessor:

BenchmarkGPT-5.2GPT-5.1Improvement
SWE-Bench Verified80.0%76.3%+3.7pt
SWE-Bench Pro55.6%50.8%+4.8pt
GPQA Diamond92.4%88.1%+4.3pt
FrontierMath T1-340.3%31.0%+9.3pt
ARC-AGI-186.2%72.8%+13.4pt
GDPval (44 occupations)70.9%38.8%+32.1pt

Notably, hallucinations are reduced by 30%, and the model maintains near-100% accuracy at 256K token context lengths.

Pricing

API pricing has changed from GPT-5.1:

GPT-5.2GPT-5.1 (reference)
Input$1.75/1M tokens$1.25/1M tokens
Cached input$0.175/1M tokens
Output$14/1M tokens$10/1M tokens

While input and output costs have increased alongside the performance gains, a substantial discount for cached inputs has been introduced.

Migration Notes

  • ChatGPT users will be automatically migrated on February 13
  • GPT-5.1 remains available as a legacy model for 3 months post-retirement
  • API is unaffected for now (separate notice will be provided)
  • A coding-optimized GPT-5.2-Codex variant is also coming soon

Sources