OpenAI readies Codex Caribou model upgrade with GPT-5.2 base

OpenAI appears to be preparing an upgrade for its Codex line, with evidence pointing to a new model internally referred to as Caribou. The latest updates to the GitHub repository for Codex reveal changes in the model selector to accommodate this addition, making it plausible that the next Codex release will be based on GPT-5.2. While the previous cycle included a Max variant, early indications suggest Caribou will launch as the primary edition without a Max version, focusing on raising the baseline capabilities.

OpenAI’s strategy with Codex has consistently targeted professional software engineering, and with GPT-5.2 setting a new benchmark in general tasks, expectations are high for a further improvement on SWE Bench Verified and similar coding leaderboards. This move is likely to benefit developers and teams using AI coding assistants, particularly those already embedded in workflows through GitHub Copilot or similar integrations. The model upgrade could also enable more advanced code generation and multi-step reasoning for enterprise and educational users.

The company’s broader product direction shows frequent iteration: following the launch of GPT-5.2, OpenAI also released its Image 1.5 model, and internal communications have referenced both Caribou and Image 1.5, though it is unclear if there is any technical linkage between the two. Given the competitive landscape, any performance gains by OpenAI in the coding space may prompt a rapid counter from Anthropic or other labs. These Codex updates are anticipated to arrive soon, possibly within the current week, with attention focused on their impact across coding benchmarks and industry adoption.