Model Comparisons

Head-to-head comparisons of AI models on real tasks.

Which Claude model to use for agents and coding. Dark editorial illustration.
Model Comparisons

Which Claude model to use for agents and coding

Sonnet 5 runs $2 per million input tokens and $10 per million output through August 31, 2026 ($3/$15 after), and it's the default executor for most coding agents now. Haiku 4.5 is $1/$5 and 4-5x faster for narrow sub-agent steps. Opus 4.8 ($5/$25) is an advisor you call in mid-task, not a full swap. Fable 5 ($10/$50) is for runs that have to keep going for hours without you.

Li Liu·Jul 17, 2026
which AI model should you point your coding agent at. Dark editorial illustration.
Model Comparisons

which AI model should you point your coding agent at

Claude Fable 5 hits 95% on SWE-bench Verified but costs $10/$50 per million tokens. Sonnet 5 scores 79.6% at $2/$10 intro pricing, roughly a fifth the cost. For most day-to-day feature work, Sonnet 5 or GPT-5.6 Terra clears the bar. Save Fable 5 or Opus 4.8 for the multi-file refactor you can't afford to get wrong.

Li Liu·Jul 17, 2026