Opus 4.6, Codex 5.3, and the post-benchmark era

kept by eddie feb 10

The latest coding models from OpenAI and Anthropic show marginal benchmark gains but real-world usability differences, with Claude ahead in product experience while Codex edges in coding capability.

GPT-5.3-Codex | Hacker News
OpenAI releases GPT-5.3-Codex, a faster agentic coding model that achieves state-of-the-art results on SWE-Bench Pro and Terminal-Bench 2.0.
also on gpt-5.3-codex, #benchmarks, OpenAI, Anthropic, #ai-coding, #developer-tools
Introducing GPT-5.3-Codex
OpenAI launches GPT-5.3-Codex, a coding agent that improves on prior benchmarks, runs 25% faster, and was used to develop itself.
also on gpt-5.3-codex, OpenAI, Anthropic, #ai-agents, #ai-coding, #developer-tools
Claude Opus 4.6
Anthropic released Claude Opus 4.6, its most capable model, scoring highest on Terminal-Bench 2.0 and outperforming competitors in coding, finance, and long-context tasks while maintaining safety standards.
also on Claude Opus 4.6, #benchmarks, Claude Code, Anthropic, #ai-agents, #ai-coding
We Tested MiniMax M2.7 Against Claude Opus 4.6 - by Darko
MiniMax M2.7 scored 56.22% on SWE-Pro, matching Claude Opus 4.6 on bug and vulnerability detection at roughly 7% of the cost in practical coding tests.
also on Claude Opus 4.6, #benchmarks, #ai-coding, #software-engineering, #developer-tools
Benchmark Scores = General Capability + Claudiness
Benchmark scores are dominated by a single 'general capability' dimension, with the second component mainly identifying a 'Claudiness' factor.
also on #benchmarks, Gemini, OpenAI, Anthropic