GPT 5 3 Codex vs Claude 4 6

Описание: Silicon Valley Sees Dual AI Launch: GPT-5.3-Codex Rivals Claude 4.6 in Programming Showdown

In a dramatic night for Silicon Valley, the competition for AI supremacy intensified as OpenAI and Anthropic released major programming AI models within hours of each other.

The sequence began with Anthropic's surprise launch of Claude Opus 4.6. In a swift countermove, OpenAI's Sam Altman announced and released GPT-5.3-Codex approximately 30 minutes later, marking a rapid escalation in the AI arms race.

OpenAI's new model, GPT-5.3-Codex, is not a general GPT-5.3 release but a specialized programming agent. It combines the top-tier coding capabilities of GPT-5.2-Codex with the advanced reasoning of GPT-5.2, boasting a 25% speed increase. It sets a new state-of-the-art (SOTA) on the SWE-Bench Pro benchmark for real-world software engineering and excels in the Terminal-Bench 2.0 evaluation.

Key advancements include its ability to handle long-horizon tasks involving research, tool use, and complex execution. It functions like a collaborative colleague, allowing real-time human guidance without losing context. Demonstrations show it can build complex games like a racing or diving game from scratch and autonomously iterate on them.

Beyond coding, GPT-5.3-Codex shows strong performance in general professional work (matching GPT-5.2 on GDPval) and computer operation tasks (OSWorld benchmark), positioning it as a step toward a universal AI agent for technical work.

The model is now available for ChatGPT paid plans across all Codex applications. Notably, OpenAI revealed it used early versions of GPT-5.3-Codex to help train and optimize its final release, accelerating its own development cycle. This launch signals Codex's evolution from a coding tool to a versatile "universal collaborator" capable of managing end-to-end workflows on a computer.
https://openai.com/index/introducing-...
https://x.com/OpenAI/status/201947415...
https://x.com/sama/status/20194747545...