GPT-5.2 is OpenAI's flagship large language model offering three variants for professional knowledge work, coding, and reasoning.
Released December 11, 2025, GPT-5.2 addresses enterprise and developer workflows with three specialized variants: Instant (fast responses), Thinking (multi-step reasoning with configurable effort), and Pro (maximum compute allocation for highest-stakes work). The model features a 400k context window, 128k output capacity, and addresses real-world gaps in long-context retrieval, agentic coding, and visual reasoning. Pricing starts at $1.75/1M input and $14/1M output tokens for Instant and Thinking, scaling to $21/$168 for Pro. While performance on science and coding benchmarks exceeds competitors, reliance on reasoning overhead increases latency for simple queries.
Bottom line: *Use for enterprise knowledge work, complex multi-step reasoning, long-context document analysis, agentic workflows, and coding tasks where quality justifies higher per-token cost; avoid for throughput-heavy applications, real-time chat, or simple classification where cheaper models suffice.*
Fast throughput variant optimized for customer support, content generation, and translation
Multi-step reasoning variant with configurable thinking effort (none/low/medium/high/xhigh)
Maximum-compute variant for high-stakes work requiring deepest reasoning and highest accuracy
View full pricing on openai.com →
Pricing may have changed since last verified. Check the official site for current plans.
No community ratings yet. Be the first to rate this tool!
No community benchmarks yet. Be the first to share a real-world data point.
No reviews yet. Be the first to share your experience.
No discussion yet. Sign in to start the conversation.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Be the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
GPT-5.2 is OpenAI’s flagship model released December 11, 2025, designed for professional knowledge work and complex problem-solving. It introduces three specialized variants—Instant, Thinking, and Pro—each allocating compute differently to balance latency and accuracy. The Instant variant optimizes for throughput in customer support and content generation. The Thinking variant introduces a reasoning dial with effort levels (none, low, medium, high, xhigh) allowing per-request trade-offs between speed and depth. Pro allocates maximum compute for high-stakes scenarios including decision support and complex planning.
The model’s 400k token context window (3x larger than GPT-5) supports 128k token outputs, enabling book-chapter-length generation and comprehensive code refactoring in single calls. Long-context retrieval maintains near-perfect accuracy even at 256k+ token depth. The Responses API /compact endpoint extends effective context for multi-turn workflows by intelligently compressing prior context rather than truncating it.
Benchmark performance demonstrates strength across reasoning, science, and coding: MMLU 92.5%, MMLU-Pro 75.4%, GPQA Diamond 92.4% (Thinking) to 93.2% (Pro), ARC-AGI-1 90.5%, AIME 2025 100%, SWE-Bench Verified 80%, SWE-Bench Pro 55.6%, and Video-MMMU 90.5%. Hallucination rates drop 30% relative to GPT-5.1 on ChatGPT’s production queries, and deception rates fall to 2.1% from o3’s 4.8%. Vision capabilities halve error rates on chart reasoning and software interface understanding.
The model maintains 90% discount on cached inputs and is available through Chat Completions API (Instant as gpt-5.2-chat-latest, Thinking as gpt-5.2) and Responses API (all variants plus Pro). A GPT-5.2-Codex variant optimizes for agentic coding. Inference performance improved ~40% in March 2026 without model weight changes. Partners including JetBrains, Augment Code, Cognition, and Databricks report measurable improvements in code review, interactive development, and data science workflows.
Share a real-world data point. Plausibility-checked by our AI moderator before publishing.
