🌥✖️💬 OpenAI Has Released GPT-5.3-Codex—Just Minutes After Anthropic's Claude Opus 4.6

Both models focus primarily on coding and agent creation. Opus 4.6 hit a record 65.4% on the Terminal-Bench 2.0 benchmark for autonomous terminal work. And just minutes later, GPT-5.3-Codex became the new leader with a score of 77.3%.

What Can They Do?

⏳ Real-world capabilities tell a better story than benchmark numbers. In 2,000 sessions, spending 2.2B tokens and $20k, Claude Opus 4.6 wrote a fully functional 100,000-line C compiler that successfully builds the Linux kernel.

💡 GPT-5.3-Codex became the first OpenAI model to build itself actively. Engineers used early versions of the AI for debugging, training, deployment, testing, and evaluating results. The team "was blown away" by how much Codex was able to accelerate its own development.

In everyday use, both models deliver impressive results—GPT-5.3-Codex and Opus 4.6 both created a 3D racing game from a single prompt. The models, textures, and physics were far from perfect, but completely functional ⤴️

Which models do you think are better?

❤️ — OpenAI
🔥 — Anthropic

@hiaimediaen