GPT-5-Codex Now Available in Cline

TL;DR

• OpenAI's GPT-5-Codex is now available in Cline — a coding-optimized version of GPT-5
• 400K context window, same pricing as GPT-5 ($1.25/M input, $10/M output), with adaptive reasoning that cuts token usage on simple tasks by 93.7%
• Update Cline and select gpt-5-codex from the model dropdown to start using it

What Dropped

OpenAI released GPT-5-Codex, a specialized variant of GPT-5 built specifically for coding agents. It's now integrated into Cline and available through the model dropdown. Unlike GPT-5's general-purpose design, Codex was trained on real-world engineering workflows: building projects, debugging, refactoring, and code review.

The Dev Angle

Adaptive reasoning changes how the model thinks. On straightforward tasks, GPT-5-Codex uses 93.7% fewer tokens than GPT-5. On complex refactoring or debugging, it extends its thinking process, iterating through implementations and test failures until it produces working code. This dynamic adjustment means you're not paying for unnecessary computation on simple requests.

It's built for agents, not chat. Codex is more steerable than general-purpose models, follows instructions precisely, and generates cleaner code without extra prompting. It's also significantly more terse than Claude — expect fewer verbose explanations and more direct output. This pairs well with Cline's thinking slider: max it out and the model allocates reasoning effort exactly where needed.

Context and pricing. The 400K context window matches GPT-5, and pricing is identical: $1.25 per million input tokens and $10 per million output tokens. If you're already using GPT-5 in Cline, the cost structure doesn't change — you're just getting a model optimized for your coding workflow.

Should You Care?

If you're running Cline with Claude Sonnet or other general-purpose models, GPT-5-Codex is worth testing. The adaptive reasoning means lower token costs on routine tasks, and the agent-specific training should produce fewer hallucinations and cleaner refactoring. The terse output takes adjustment if you're used to Claude's explanatory style, but that's actually a feature for agent workflows.

If you're already on GPT-5, this is a straightforward upgrade — same price, better coding performance. If you're cost-conscious and mostly doing simple edits, the 93.7% token savings on basic tasks could add up. If you need general-purpose reasoning beyond coding, stick with GPT-5.

Update Cline and select gpt-5-codex from your provider settings to get started.

Source: Cline