1 min read

AI Race Heats Up: GPT-5.3 Builds Itself as Opus 4.6 Compiles a C Compiler

New AI Models

OpenAI Releases GPT-5.3-Codex with Recursive Self-Improvement Capabilities
OpenAI has launched GPT-5.3-Codex, a groundbreaking AI model capable of recursive self-improvement, where early versions were used to debug its own training, manage deployment, and diagnose test results. The model excels in long-running tasks, maintaining context for complex research, tool use, and execution.


Anthropic Unveils Claude Opus 4.6 with Major Benchmark Improvements
Anthropic’s Claude Opus 4.6 introduces significant performance upgrades, nearly doubling its ARC-AGI 2 benchmark score compared to Opus 4.5, while maintaining the same pricing. The model is optimized for ambitious, long-running tasks, as demonstrated by its ability to autonomously build a C compiler that works with the Linux kernel.


AI Development Tools & Platforms

Perplexity Launches Model Council for Enhanced Multi-LLM Reasoning
Perplexity AI has introduced the Perplexity Model Council, a feature for Perplexity Max users that delegates tasks to a "swarm" of frontier LLMs. These models work asynchronously to provide deeper, multi-perspective answers, improving response accuracy and depth.