Particle.news
Download on the App Store

OpenAI Launches GPT-5.2, Rolling Out to Paid ChatGPT Tiers and the API

OpenAI frames the upgrade as a push to counter Google’s Gemini 3 with expert-level performance on real tasks.

Overview

  • The phased release begins today for Plus, Pro, Business, Enterprise and Go subscribers, with developer access via the API and broader availability for free users to follow.
  • GPT-5.2 comes in three modes—Instant for speed, Thinking for deeper reasoning and Pro for highest-quality output—targeting professional knowledge work and long‑running agents.
  • OpenAI reports benchmark gains for GPT-5.2 Thinking, including 70.9% on GDPval versus top professionals, 55.6% on SWE‑Bench Pro, about 92% on GPQA Diamond and perfect scores on some AIME 2025 tests.
  • The company says inaccurate responses drop by roughly 30% versus GPT‑5.1, with stronger long‑context handling (up to a reported 400,000 tokens), more reliable tool use and better vision for charts, dashboards and UI screenshots.
  • Coverage notes the launch follows an internal “code red” and compares results with Gemini 3 and Claude, while OpenAI keeps legacy models available during a transition and prices the API higher than GPT‑5.1 but below rival frontier tiers.