GPT-5.2 Arrives: What Changed for Developers?
OpenAI released GPT-5.2 with a 40% price increase and expanded context windows, while quietly rolling out Skills support in ChatGPT and Codex CLI—a lightweight pattern for extending model capabilities.
OpenAI released GPT-5.2 on December 11, bringing expanded context windows and enhanced reasoning capabilities to its flagship model. But the more significant development for developers may be what arrived alongside it: Skills support, now quietly available in both ChatGPT and the company's open-source Codex CLI tool.
What Changed in GPT-5.2
GPT-5.2 expands the model's context window to 256,000 tokens—roughly 200 pages of documents—enabling it to handle long-form workflows that previously required chunking or summarization. According to the company's benchmarks, GPT-5.2 Thinking achieved near-perfect accuracy on the four-needle MRCR evaluation at the full 256,000-token context window.
The model also brings improvements to multimodal vision and spatial reasoning, making it more capable at analyzing dashboards, diagrams, and complex documents. Enhanced tool-using capabilities through the Responses API give developers better integration options for building data-aware agentic systems.
But these improvements come at a cost. API pricing now sits at $1.75 per million input tokens and $14 per million output tokens—a 40% increase over GPT-5.1, according to the official pricing page. For developers building cost-sensitive applications, that price jump requires careful consideration of whether the performance gains justify the additional spend.
Skills: The Quiet Revolution
While GPT-5.2 grabbed headlines, the addition of Skills support represents a more fundamental shift in how developers can extend model capabilities. As developer Simon Willison discovered, ChatGPT's Code Interpreter now includes a /home/oai/skills folder accessible through simple prompts.
Skills are folders containing a markdown file and optional scripts or resources. The pattern originated with Anthropic in October 2025, and OpenAI's rapid adoption—just two months later—signals broad industry acceptance of the approach.
"One of the things that most excited me about Anthropic's new Skills mechanism back in October is how easy it looked for other platforms to implement," Willison wrote on his blog. "It turns out OpenAI are doing exactly that."
The simplicity is deliberate. A skill is just structured documentation that tells the model how to perform specific tasks. OpenAI's current skills cover spreadsheets, DOCX files, and PDFs. Notably, their PDF handling converts documents to per-page PNGs and processes them through vision-enabled models—preserving layout and graphics information that text extraction alone would miss.
Willison tested the feature by asking ChatGPT to create a PDF about rimu trees and kākāpō breeding season. The model started by "reading skill.md for PDF creation guidelines," then spent over eleven minutes producing a detailed document. At one point during generation, it detected that the font didn't support macrons in "kākāpō" and switched to an alternative.
Skills in Codex CLI
The Skills implementation extends beyond ChatGPT. OpenAI's Codex CLI—an open-source coding agent with more than 52,000 stars on GitHub—added experimental Skills support two weeks before the GPT-5.2 announcement, according to the project's commit history.
Any folder placed in ~/.codex/skills becomes available to the tool when launched with the --enable skills flag. Developers can list available skills and invoke them through natural language prompts.
Willison demonstrated this by creating a Datasette plugin skill using Claude, then installing it in Codex CLI. With a simple prompt—"Write a Datasette plugin in this folder adding a /-/cowsay?text=hello page that displays a pre with cowsay from PyPI saying that text"—the tool generated working plugin code.
The lightweight nature of skills makes them easy to share and remix. Unlike complex tool integrations or API-specific implementations, a skill folder can work across different platforms that support the pattern.
Why This Matters
The Skills pattern addresses a persistent challenge in AI development: how to extend model capabilities without building custom integrations for each use case. Rather than hard-coding specific behaviors into prompts or building elaborate tool chains, developers can package expertise into portable folders.
The rapid adoption across both Anthropic and OpenAI suggests the pattern fills a real need. As Willison noted, "When I first wrote about skills in October I said Claude Skills are awesome, maybe a bigger deal than MCP. The fact that it's just turned December and OpenAI have already leaned into them in a big way reinforces to me that I called that one correctly."
The informal nature of the specification—essentially just a convention for folder structure and markdown files—may benefit from standardization. Willison suggested the newly formed Agentic AI Foundation, which includes OpenAI and Anthropic as members, could take this on. The foundation, established under the Linux Foundation, aims to provide neutral governance for the agentic AI ecosystem.
Integration Considerations
For developers building on OpenAI's platform, the GPT-5.2 release presents both opportunities and constraints. The expanded context window opens up use cases previously impractical—full document analysis, multi-file codebases, extensive conversation history—but the 40% price increase means cost modeling needs revision.
The Responses API, now available with GPT-5.2, provides structured endpoints for building agents with tool use, reasoning, and multimodal capabilities. Platform integrations from Databricks and Snowflake indicate enterprise adoption is already underway.
Skills support in Codex CLI is marked as experimental, which means the interface may change. Developers exploring the feature should expect iteration. The documentation lives in the project's docs/skills.md file on GitHub.
What Developers Should Do
If you're currently using GPT models in production, audit your token usage against the new pricing. The 40% increase compounds quickly at scale. For applications that don't require the expanded context window or enhanced reasoning, GPT-5.1 remains available at the lower price point.
Experiment with Skills if your workflow involves repetitive specialized tasks. The pattern works best when you can codify expertise into reusable instructions. Check the Anthropic skills repository and OpenAI's ChatGPT implementation for examples of how different organizations structure their skills.
For CLI-based development workflows, Codex CLI with Skills support offers a way to extend the tool's capabilities without waiting for official feature releases. Install it via npm (npm install -g @openai/codex) or Homebrew (brew install --cask codex) and explore the skills directory.
The Bottom Line
GPT-5.2 delivers meaningful improvements in context handling and reasoning, but at a significantly higher price. The more interesting story is Skills—a lightweight, portable pattern for extending model capabilities that's gaining rapid cross-platform adoption. As the ecosystem converges on shared patterns like Skills and the Model Context Protocol, developers gain more flexibility in how they build and deploy AI features.
The question isn't whether to adopt GPT-5.2—that depends on your specific use case and budget. The question is whether you're positioning your development workflow to take advantage of emerging patterns that work across multiple AI platforms. Skills support represents that kind of foundational shift.