Anthropic and OpenAI just dropped their new flagship coding models.
Same day. Same target.
2 min read
Originally posted on LinkedIn · 27 reactions · 1 comments · View original →
- Anthropic and OpenAI just dropped their new flagship coding models. Within one hour of each other.
Same day. Same target. Same message: "We're coming for your developers."
Opus 4.6:
- 1M token context (first for Opus-class)
- Context rot mostly solved
- Agent teams: parallel sub-agents that coordinate autonomously
Codex 5.3:
- First model that was "instrumental in creating itself"
- OSWorld jumps from 38% to 65%
- 25% faster inference
This is their initial response to the OpenClaw phenomenon.
Both labs are pivoting hard toward autonomous agency. "Labour as a Service" is the framing now. Not chatbots. Not copilots. AI employees.
What their testimonials are saying:
Opus 4.6: "Autonomously closed 13 issues and assigned 12 to the right team members in a single day, managing ~50-person org across 6 repos."
Codex 5.3: "Handled a multi-million-line codebase migration like a senior engineer. Planned. Adapted. Finished in half the time."
That's not copilot territory. That's replacement territory.
To make it even more interesting, Anthropic ran Super Bowl ads mocking ads inside AI responses.
They never named OpenAI. But everyone knew.
Sam Altman responded with an essay on X. The rebuttal got over 10M views. The original ad? 2.7M. 😅
Today I’ll be testing the new agent team's capability of Opus 4.6. It looks super powerful and aligned with the principle of having humans in the loop stearing rather than push-and-pray.