Opus 5: The first practical default model for AI-assisted coding

Anthropic dropped Opus 5, and it's immediately competitive with top-tier models at roughly half the API price. The creator reports strong real-world coding results, particularly around merge-ready code and diligent verification behavior. Opus 5 consistently scores well on code benchmarks like Frontier Code and agentic coding tasks, often beating Fable in efficiency per dollar. The

key claim is that Opus sits in a valuable middle ground. It combines the instruction-following eagerness of OpenAI's models with the code quality and taste that Anthropic's Fable is known for. The creator found it uses subscription limits much more slowly than Fable, making it a better economic default. However, this advantage is partly due

to very different token utilization, not raw token cost. Important tradeoffs exist. Opus 5 uses more tokens per task than Fable, making cost savings closer to 20-25% than the advertised 50% in real use. It can be slower and over-verify its work. The factual knowledge is slightly behind Fable, leading to occasional confident fabrications. Safety

alignment is baked in, unlike Fable's classifier-only approach, which opens it for enterprise use. For engineering teams, the actionable insight is clear: this model reduces the need to context-switch between Fable for planning and cheap models for execution. It is a strong daily driver candidate. Teams should trial it on their actual workflows rather than