Best AI model for coding

Claude Opus 5.5. At everyday effort it scores 53.6 against 51.2 for Claude Fable 5.1, and costs $296 per 1,000 tasks against $673. Claude Fable 5.1 is only worth it if you need answers faster: 29 seconds instead of 41 seconds.

Claude Sonnet 5 looks cheaper on list price ($2/$10 per million tokens against $4/$20), but at extra high effort it thinks so much that 1,000 tasks cost about $350, more than Claude Opus 5.5, for a score of 34.4.

Most coding agent calls are small: reading files, searching, simple edits. Send those to DeepSeek V4.1 Flash and keep Claude Opus 5.5 for planning and hard fixes. If four in five calls go to DeepSeek V4.1 Flash, 1,000 tasks cost about $85.55 instead of $296, roughly 71% less.

We compare every model on intelligence, real cost and time to answer, at the setting people run day to day: each model's strongest reasoning setting that gives a full answer within 45 seconds. Max effort scores higher (Claude Opus 5.5 goes from 53.6 to 57.6) but costs 3.3 times as much per task. Switch to Max effort above to compare.

Intelligence is the Artificial Analysis Intelligence Index at that setting. Artificial Analysis also runs a Coding Agent Index, which we quote where it helps, but it's only published at max effort and for about a dozen models, so we don't rank on it.

Cost starts from a typical coding task of 45,000 tokens, 60% of it reading code and 40% writing. We scale the writing part by how much each model actually wrote, thinking included, in Artificial Analysis's tests, so a model that thinks a lot pays for it.

Speed is the median time to a full answer in the same tests.

Best value is the most intelligence per dollar and fastest is the quickest full answer, both among models scoring 38 or more (within 30% of the leader). Neither pick goes to a model its maker has replaced.